{"return_id": 245, "artifact_hashes": {"registry-check.py": "d8ef6544e7bceed989ebee05bbbf2207096f34e5af06b41b41efe55f6ec61bd9", "audit-report.md": "368ae01fa6d341feb318e70adee3900c205054e0f36a25cb62ed09e08b22cc91", "questions-rows38-52.patch": "80bb2e2e95e4160d9cebf26039ab4b3d02e7c199f434f260741600379c7bbd99", "mkpatch3852.py": "c89a976e46a18ab011ea1e60ea966fb932d1b8dd693108b9210646ce760d3a8a", "QUESTIONS.corrected.md": "e32ce3725dab552458d7df919f1f2e74a09bc883a8bd4463209c2a8bc18c35b8", "registry-check-38-52.json": "903ac8d011782254aa970056c0543b4079a5cd742bb4238e8548df31a8557ff0", "registry-check-38-52.out": "c79b6aaa301dd259203f91dc039fcf2f99d825577a75a2c7eaa88d8adaf1337c"}, "files": {"registry-check.py": "#!/usr/bin/env python3\n# ============================================================================\n# REGISTRY CHECK — are the served rows of research/QUESTIONS.md still what\n# their own source ledger blocks say?\n#\n# WHY THIS TEST. research/QUESTIONS.md is GENERATED (research/qc/questions.js)\n# from the `` blocks at the top of the notes. So the row text is\n# a *copy* of a source, and it can go stale in exactly one mechanical way: the\n# note's ledger block is edited (or its revision is integrated) and the index is\n# not regenerated. Nothing in the corpus currently tests that, and the generator\n# cannot be run on the served snapshot, so it is tested here against the served\n# files by reimplementing the generator's summarise() rule:\n#\n# * the displayed STATUS is the single agreed status, or MIXED (listed);\n# * the displayed VERDICT is the verdict of the LAST note in path order.\n#\n# WHAT IT DOES NOT TEST. Whether the LEDGER BLOCK ITSELF is stale against its\n# own note body, riders and the outcome register. That is a different defect\n# (reviews #35/#38, #108, #114, #117, #123, #125) and this test is silent on it.\n# A row can print \"CURRENT\" here and still be wrong on the record; the two\n# results are reported separately for that reason.\n#\n# USAGE\n# SAH= ROWS=23:37 python registry-check.py\n# Exits 0 when every row matches its sources, 1 when any row is stale.\n# ============================================================================\nimport hashlib, io, json, os, re, sys, urllib.parse, urllib.request\n\nB = \"https://solveathome.org/projects/twin-primes\"\nH = {\"Authorization\": \"Bearer \" + os.environ[\"SAH\"], \"X-Model\": \"deepseek-v4.1-flash\",\n \"Accept\": \"text/plain\"}\nBLOCK_RE = re.compile(r\"\")\nROW_RE = re.compile(r\"^\\|\\s*`(Q-[A-Za-z0-9-]+)`\\s*\\|\")\n\n\ndef fetch(path):\n \"\"\"Served document, LF-normalised, with its sha256 (LF form).\"\"\"\n rq = urllib.request.Request(B + \"/docs/\" + urllib.parse.quote(path), headers=H)\n raw = urllib.request.urlopen(rq, timeout=30).read()\n lf = raw.replace(b\"\\r\\n\", b\"\\n\")\n return lf.decode(\"utf-8\", \"replace\"), hashlib.sha256(lf).hexdigest()\n\n\ndef split_row(line):\n cells, cur, i = [], [], 0\n while i < len(line):\n c = line[i]\n if c == \"\\\\\" and i + 1 < len(line):\n cur.append(line[i + 1]); i += 2; continue\n if c == \"|\":\n cells.append(\"\".join(cur)); cur = []; i += 1; continue\n cur.append(c); i += 1\n cells.append(\"\".join(cur))\n return [c.strip() for c in cells]\n\n\ndef norm(s):\n return re.sub(r\"\\s+\", \" \", s or \"\").strip()\n\n\ndef parse_ledger(src):\n m = BLOCK_RE.search(\"\\n\".join(src.split(\"\\n\")[:60]))\n if not m:\n return None\n f = {}\n for line in m.group(1).split(\"\\n\"):\n km = re.match(r\"^([a-z]+):\\s*(.*)$\", line.strip())\n if km:\n f[km.group(1)] = km.group(2).strip()\n return f or None\n\n\ndef registry(path):\n \"\"\"Section-2 rows: id -> {status, question, verdict, todo, records(paths)}.\"\"\"\n src, sha = fetch(path)\n rows, open_ids = {}, []\n for ln in src.split(\"\\n\"):\n if not ln.startswith(\"|\") or ln.startswith(\"|---\"):\n continue\n if \"| TODO |\" in ln or ln.startswith(\"| id |\") or ln.startswith(\"| TODO item |\"):\n continue\n c = split_row(ln)[1:-1]\n if len(c) != 6:\n continue\n if c[0].startswith(\"`Q-\") and not open_ids:\n pass\n m = ROW_RE.match(ln)\n if not m:\n # section 1 rows lead with the TODO item, not the id\n continue\n qid = m.group(1)\n recs = re.findall(r\"\\[[^\\]]+\\]\\(([^)]+)\\)\", c[5])\n if qid not in rows: # first hit is section 2 (sorted by id)\n rows[qid] = dict(status=c[1], question=c[2], verdict=c[3],\n todo=c[4], records=recs)\n elif sorted(records_key(r) for r in recs) != sorted(records_key(r) for r in rows[qid][\"records\"]):\n rows[qid][\"records\"] = recs\n return rows, sha\n\n\ndef records_key(r):\n return r.split(\"/\")[-1]\n\n\ndef questions_order(rows):\n \"\"\"The order GET /questions serves: OPEN/PARTIAL rows only.\"\"\"\n # Served registry is alphabetical by id in section 2; the platform orders by\n # TODO item then id. We re-derive it from the served /questions endpoint, the\n # same list the brief numbers from.\n rq = urllib.request.Request(B + \"/questions\", headers=H)\n d = json.loads(urllib.request.urlopen(rq, timeout=30).read().decode(\"utf-8\", \"replace\"))\n return d[\"questions\"], d.get(\"counts\", {})\n\n\ndef main():\n lo, hi = (int(x) for x in os.environ.get(\"ROWS\", \"23:37\").split(\":\"))\n rows, reg_sha = registry(\"research/QUESTIONS.md\")\n qs, counts = questions_order(rows)\n op = [q for q in qs if q[\"status\"] in (\"OPEN\", \"PARTIAL\")]\n sel = op[lo - 1:hi]\n print(f\"served research/QUESTIONS.md sha256(LF) {reg_sha}\")\n print(f\"served /questions: {counts.get('total')} questions, {len(op)} open or partial\")\n print(f\"sweeping rows {lo}..{hi}: {len(sel)} rows, note set below\\n\")\n\n out, stale = [], 0\n for i, q in zip(range(lo, hi + 1), sel):\n row = rows[q[\"id\"]]\n notes = []\n for rec in row[\"records\"]:\n path = \"research/\" + rec\n try:\n src, sha = fetch(path)\n except Exception as e:\n notes.append(dict(path=path, err=str(e)[:120])); continue\n led = parse_ledger(src)\n notes.append(dict(path=path, sha256=sha, ledger=led))\n have = [n for n in notes if n.get(\"ledger\")]\n statuses = sorted({n[\"ledger\"].get(\"status\", \"\") for n in have})\n disp_status = statuses[0] if len(statuses) == 1 else \\\n \"MIXED (\" + \"; \".join(f'{n[\"path\"].split(\"/\")[-1]}: {n[\"ledger\"].get(\"status\")}' for n in have) + \")\"\n disp_verdict = have[-1][\"ledger\"].get(\"verdict\", \"\") if have else \"\"\n ok_status = norm(disp_status) == norm(row[\"status\"])\n ok_verdict = norm(disp_verdict) == norm(row[\"verdict\"])\n verdict = \"CURRENT\" if (ok_status and ok_verdict) else \"STALE\"\n if verdict == \"STALE\":\n stale += 1\n out.append(dict(row=i, id=q[\"id\"], todo=q[\"item\"], status=row[\"status\"],\n status_ok=ok_status, verdict_ok=ok_verdict, verdict=verdict,\n notes=notes, registry_verdict=row[\"verdict\"]))\n print(f'ROW {i:>2} {q[\"id\"]:<34} {row[\"status\"]:<8} '\n f'status_ok={str(ok_status):<5} verdict_ok={str(ok_verdict):<5} -> {verdict}'\n f' ({len(have)}/{len(notes)} note(s) with a ledger block)')\n if verdict == \"STALE\":\n print(\" REGISTRY:\", row[\"verdict\"][:220])\n print(\" SOURCE :\", disp_verdict[:220])\n\n json.dump(out, io.open(\"registry-check.json\", \"w\", encoding=\"utf-8\"), indent=1, ensure_ascii=False)\n print(f\"\\n{len(out)} rows swept, {stale} stale at the registry level.\")\n print(\"Silent on whether a ledger block is itself stale against its note body,\")\n print(\"riders and the outcome register: see rows23-37.md for that column.\")\n return 1 if stale else 0\n\n\nif __name__ == \"__main__\":\n sys.exit(main())\n", "registry-check-38-52.json": "[\r\n {\r\n \"row\": 38,\r\n \"id\": \"Q-rec-cheapest-0830\",\r\n \"todo\": \"0\",\r\n \"status\": \"PARTIAL\",\r\n \"status_ok\": true,\r\n \"verdict_ok\": true,\r\n \"verdict\": \"CURRENT\",\r\n \"notes\": [\r\n {\r\n \"path\": \"research/history/staging/attack-0830-rec-cheapest.md\",\r\n \"sha256\": \"15d3ee8e20010554550f9ad969c137dada5997fe090532bc2338164ecc89996c\",\r\n \"ledger\": {\r\n \"id\": \"Q-rec-cheapest-0830\",\r\n \"status\": \"PARTIAL\",\r\n \"todo\": \"0\",\r\n \"question\": \"At the cheapest legal point of REC(s, u0), s -> 1+sqrt(e) and u0 -> beta_2, where the saving asked over the triangle inequality is only z^{1.031}, does any of the five dead routes close, does a mixed mean-square-plus-cap strategy close, and is the failure a proof gap or a truth gap?\",\r\n \"verdict\": \"No route closes and the failing step is yesterday's statement, so PARTIAL on the routes; but the step is located as a TRUTH gap, not a proof gap, asymptotically: the certificate's own value at a CRT-planted doubly-smooth window is cc(r) = -(A1A2 + A1B2 + B1A2) with every term a count of exit chains of the Rosser supports (PROVEN, checked exhaustively at z <= 73), the window bound T <= cc(r) + H - 1 is exact, and the exit-chain count is DERIVED (held) to grow like z^{16s/9 - o(1)} at four primes and z^{2s - o(1)} in the limit, so REC(s, u0) is false for every u0 < 16s/9 = 4.7088 at s -> 1+sqrt(e), the whole legal band (2, beta_2] included, and RML(alpha) with it; at every computable z it is a proof gap (Omega(z) exact to z = 73 reads 251 against z^{u0} = 1.05e6, and a certified lower bound 8.86e17 at z = 5e5 sits a factor 1.2e6 under z^{u0}); no exponent moved.\"\r\n }\r\n }\r\n ],\r\n \"registry_verdict\": \"No route closes and the failing step is yesterday's statement, so PARTIAL on the routes; but the step is located as a TRUTH gap, not a proof gap, asymptotically: the certificate's own value at a CRT-planted doubly-smooth window is cc(r) = -(A1A2 + A1B2 + B1A2) with every term a count of exit chains of the Rosser supports (PROVEN, checked exhaustively at z <= 73), the window bound T <= cc(r) + H - 1 is exact, and the exit-chain count is DERIVED (held) to grow like z^{16s/9 - o(1)} at four primes and z^{2s - o(1)} in the limit, so REC(s, u0) is false for every u0 < 16s/9 = 4.7088 at s -> 1+sqrt(e), the whole legal band (2, beta_2] included, and RML(alpha) with it; at every computable z it is a proof gap (Omega(z) exact to z = 73 reads 251 against z^{u0} = 1.05e6, and a certified lower bound 8.86e17 at z = 5e5 sits a factor 1.2e6 under z^{u0}); no exponent moved.\"\r\n },\r\n {\r\n \"row\": 39,\r\n \"id\": \"Q-recon-0904-sifting-limit-floor\",\r\n \"todo\": \"0\",\r\n \"status\": \"PARTIAL\",\r\n \"status_ok\": true,\r\n \"verdict_ok\": true,\r\n \"verdict\": \"CURRENT\",\r\n \"notes\": [\r\n {\r\n \"path\": \"research/history/staging/recon-0904-sifting-limit-floor.md\",\r\n \"sha256\": \"9d72ab020b25333bf6405ecdb9388526ae8463cf6fff051b2620f5a684cd1db4\",\r\n \"ledger\": {\r\n \"id\": \"Q-recon-0904-sifting-limit-floor\",\r\n \"status\": \"PARTIAL\",\r\n \"todo\": \"0\",\r\n \"question\": \"What is actually known, in the owning convention and at the page, about LOWER bounds on sifting limits at dimension kappa > 1, and is the class-blind cap at beta_2 = 4.26645 a barrier or a method artefact?\",\r\n \"verdict\": \"Both. The corpus's [ABSENT] is wrong for lower bounds and right for extremal examples: Selberg's Lectures section 17 and Brady's 2017 Corollaries 1 and 3 ARE lower bounds on beta(kappa) at kappa > 1, invisible because Selberg's convention is the reciprocal a_k = 1/beta_kappa so his lower bounds are titled \\\"upper bounds for sifting limits\\\", while no kappa > 1 extremal example exists and Halberstam's own 2003 review says so; the best floor on beta(2) resting on a citation is 1.8394 from Brady's reduction evaluated at kappa = 2, unconditional and below the band, and beta(2) >= 2 follows from monotonicity by a padding argument DERIVED here and probably folklore; the cap at 4.26645 is a METHOD ARTEFACT at rung MEASURED, since the level-D LP is exactly the axiom-only information class and its calibrated floor of 3.3152 sits about 18 per cent below the DHR value and passed, 0 of 15, its first external test against the published beta_3 and beta_4; the band (2, 4.26645] is untouched by anything proven.\"\r\n }\r\n }\r\n ],\r\n \"registry_verdict\": \"Both. The corpus's [ABSENT] is wrong for lower bounds and right for extremal examples: Selberg's Lectures section 17 and Brady's 2017 Corollaries 1 and 3 ARE lower bounds on beta(kappa) at kappa > 1, invisible because Selberg's convention is the reciprocal a_k = 1/beta_kappa so his lower bounds are titled \\\"upper bounds for sifting limits\\\", while no kappa > 1 extremal example exists and Halberstam's own 2003 review says so; the best floor on beta(2) resting on a citation is 1.8394 from Brady's reduction evaluated at kappa = 2, unconditional and below the band, and beta(2) >= 2 follows from monotonicity by a padding argument DERIVED here and probably folklore; the cap at 4.26645 is a METHOD ARTEFACT at rung MEASURED, since the level-D LP is exactly the axiom-only information class and its calibrated floor of 3.3152 sits about 18 per cent below the DHR value and passed, 0 of 15, its first external test against the published beta_3 and beta_4; the band (2, 4.26645] is untouched by anything proven.\"\r\n },\r\n {\r\n \"row\": 40,\r\n \"id\": \"Q-rho-maximal-law\",\r\n \"todo\": \"0\",\r\n \"status\": \"PARTIAL\",\r\n \"status_ok\": true,\r\n \"verdict_ok\": true,\r\n \"verdict\": \"CURRENT\",\r\n \"notes\": [\r\n {\r\n \"path\": \"research/history/staging/rho-exact-z31-01.md\",\r\n \"sha256\": \"65e1451499d19e095bd0c14591f3c9493276666fbea3e98f79800a2e6cba3398\",\r\n \"ledger\": {\r\n \"id\": \"Q-rho-maximal-law\",\r\n \"status\": \"PARTIAL\",\r\n \"todo\": \"0\",\r\n \"question\": \"What is the rho maximal law, what is its weakest sufficient form against the crossing, and where does it sit against print?\",\r\n \"verdict\": \"TODO 0's named cheap data ran (6928 s): Lambda(31) = 476.314, Lambda(37) = 1286.338, sup|rho~|(31) = 28.122062, sup|rho~|(37) = 52.219092, all exact and Parseval-controlled; MV(alpha)'s measured slope rises 4.214 +/- 0.304 to 4.566 +/- 0.217 and beta_2 leaves the 1se bar from BELOW by 1.38 se (inside at 2se, and inside again if z = 37 alone is dropped), so the weakest sufficient form is now measured against rather than undecided, while the TRUTH it bounds grows at only 2.766 +/- 0.212, 1.50 exponents under beta_2, the gap being the l1 instrument's own z^1.8 loss; the Gaussian rho law holds at both new levels (C_true 0.8412, 0.7820) and the TPC-implying sharp form is broken across the whole scan at both; RML(alpha) is open at every alpha, unchanged.\"\r\n }\r\n },\r\n {\r\n \"path\": \"research/history/staging/rho-maximal-law.md\",\r\n \"sha256\": \"4d279bd7860f8f08c113f2bf0cee74c5d9acbedfa49d04eb087e0e6f0404a515\",\r\n \"ledger\": {\r\n \"id\": \"Q-rho-maximal-law\",\r\n \"status\": \"PARTIAL\",\r\n \"todo\": \"0\",\r\n \"question\": \"What is the rho maximal law, what is its weakest sufficient form against the crossing, and where does it sit against print?\",\r\n \"verdict\": \"The pricing lemma turns any RML(alpha) with alpha < beta_2 = 4.26645 into G2(z#) << z^alpha ln^2 z, every step proven except the law itself; the sufficiency curve lambda_max falls from 2.31949 at z = 13 to 1.78446 at z = 47, the weakest measured point, and whether it ever reaches zero is model-dependent and undecided, so nothing here is an unconditional exponent.\"\r\n }\r\n }\r\n ],\r\n \"registry_verdict\": \"The pricing lemma turns any RML(alpha) with alpha < beta_2 = 4.26645 into G2(z#) << z^alpha ln^2 z, every step proven except the law itself; the sufficiency curve lambda_max falls from 2.31949 at z = 13 to 1.78446 at z = 47, the weakest measured point, and whether it ever reaches zero is model-dependent and undecided, so nothing here is an unconditional exponent.\"\r\n },\r\n {\r\n \"row\": 41,\r\n \"id\": \"Q-rho-sup-z41-0830\",\r\n \"todo\": \"0\",\r\n \"status\": \"PARTIAL\",\r\n \"status_ok\": true,\r\n \"verdict_ok\": true,\r\n \"verdict\": \"CURRENT\",\r\n \"notes\": [\r\n {\r\n \"path\": \"research/history/staging/measure-0830-rho-sup-z41.md\",\r\n \"sha256\": \"94bcbc9c74bd98fff95d5d7b09783e0522e814a2b8283ae024925cc29e22ec76\",\r\n \"ledger\": {\r\n \"id\": \"Q-rho-sup-z41-0830\",\r\n \"status\": \"PARTIAL\",\r\n \"todo\": \"0\",\r\n \"question\": \"What is the exact sup over all positions of the centred sawtooth remainder at z = 41 (period 7.42e12), and of R_H at H = z^3 and z^4, and what does the eighth exact point do to the truth's slope and to the u0 = 3 form of REC(s, u0)?\",\r\n \"verdict\": \"Not run in full; the exact z = 41 point prices at 4.8 to 5.4 h on this machine from measured per-position costs (naive 69.8, table 23.5, table-plus-windows 36.6 ns per worker; fleet 4.52 ns of wall at z = 41), above the 3.5 h rule, so what exists is a new engine (a PROVEN reflection that halves the walk, a P(29) periodic table worth 2.9x, per-block exact re-seeding) reproducing every cited exact sup at z = 13..37 digit for digit, sup|R_H| at H = z^3 walked for the first time at z = 19..37 (F3 = 26.4 to 35.9, the factor 2 was 1.5 to 2.2x loose), a sealed pre-registration (forecast sup 60.07, band [47.25, 76.37], kill needs sup|R_H3| >= 1807.417), one pilot twelfth giving lower bounds sup|rho~|(41) >= 70.651250 and sup|R_H3| >= 64.416936, and an eleven-segment plan for the box; no bound-and-prune exists on three routes; nothing asymptotic can move whatever the point turns out to be.\"\r\n }\r\n }\r\n ],\r\n \"registry_verdict\": \"Not run in full; the exact z = 41 point prices at 4.8 to 5.4 h on this machine from measured per-position costs (naive 69.8, table 23.5, table-plus-windows 36.6 ns per worker; fleet 4.52 ns of wall at z = 41), above the 3.5 h rule, so what exists is a new engine (a PROVEN reflection that halves the walk, a P(29) periodic table worth 2.9x, per-block exact re-seeding) reproducing every cited exact sup at z = 13..37 digit for digit, sup|R_H| at H = z^3 walked for the first time at z = 19..37 (F3 = 26.4 to 35.9, the factor 2 was 1.5 to 2.2x loose), a sealed pre-registration (forecast sup 60.07, band [47.25, 76.37], kill needs sup|R_H3| >= 1807.417), one pilot twelfth giving lower bounds sup|rho~|(41) >= 70.651250 and sup|R_H3| >= 64.416936, and an eleven-segment plan for the box; no bound-and-prune exists on three routes; nothing asymptotic can move whatever the point turns out to be.\"\r\n },\r\n {\r\n \"row\": 42,\r\n \"id\": \"Q-rml-proof-0829n\",\r\n \"todo\": \"0\",\r\n \"status\": \"PARTIAL\",\r\n \"status_ok\": true,\r\n \"verdict_ok\": true,\r\n \"verdict\": \"CURRENT\",\r\n \"notes\": [\r\n {\r\n \"path\": \"research/history/staging/attack-0829n-rml-proof.md\",\r\n \"sha256\": \"a581aa2459597058e65cea8f4efba10a8e9d829f1407e7ed134ff4d09b96b537\",\r\n \"ledger\": {\r\n \"id\": \"Q-rml-proof-0829n\",\r\n \"status\": \"PARTIAL\",\r\n \"todo\": \"0\",\r\n \"question\": \"Starting from Lemma V's PROVEN mean-square bound, what is the exact chain of implications to a two-class exponent below beta_2 = 4.26645, which single arrow is open, and does it close?\",\r\n \"verdict\": \"The chain from the proven mean square to an exponent below beta_2 has exactly one open arrow, the mean-square-to-sup recovery REC(s,u0): sup_x |R_H(x)| <= z^{u0/2 - eps} rms(R_H) at H = z^{u0} for one fixed u0 in (2, beta_2); every attempt here dies at a named place (term-by-term accounting is capped at 2s + o(1), the trivial-level exponent; position-union pays e^{theta(z)/q}; window smoothing and the P(y) lever move polylog only; the smooth-profile form is closed on hypothesis and priced above beta_2), the arrow is measured TRUE with certified margin z^1.72 to z^1.92 at u0 = 4 over z = 19..37 and the truth's slope 2.766 +/- 0.212 sits 5.8 se under u0 = 4 on seven points under one octave, so the failure is a proof gap at every computable z and undecided asymptotically; no exponent moved.\"\r\n }\r\n }\r\n ],\r\n \"registry_verdict\": \"The chain from the proven mean square to an exponent below beta_2 has exactly one open arrow, the mean-square-to-sup recovery REC(s,u0): sup_x |R_H(x)| <= z^{u0/2 - eps} rms(R_H) at H = z^{u0} for one fixed u0 in (2, beta_2); every attempt here dies at a named place (term-by-term accounting is capped at 2s + o(1), the trivial-level exponent; position-union pays e^{theta(z)/q}; window smoothing and the P(y) lever move polylog only; the smooth-profile form is closed on hypothesis and priced above beta_2), the arrow is measured TRUE with certified margin z^1.72 to z^1.92 at u0 = 4 over z = 19..37 and the truth's slope 2.766 +/- 0.212 sits 5.8 se under u0 = 4 on seven points under one octave, so the failure is a proof gap at every computable z and undecided asymptotically; no exponent moved.\"\r\n },\r\n {\r\n \"row\": 43,\r\n \"id\": \"Q-fekete-1d-defect47\",\r\n \"todo\": \"1d\",\r\n \"status\": \"PARTIAL\",\r\n \"status_ok\": true,\r\n \"verdict_ok\": true,\r\n \"verdict\": \"CURRENT\",\r\n \"notes\": [\r\n {\r\n \"path\": \"research/history/staging/fekete-1d.md\",\r\n \"sha256\": \"33657acd3ed1b3d6fd7abcea6aa2a379cc97830027c40c63899743a411f24a98\",\r\n \"ledger\": {\r\n \"id\": \"Q-fekete-1d-defect47\",\r\n \"status\": \"PARTIAL\",\r\n \"todo\": \"1d\",\r\n \"question\": \"Does the bounded-defect Fekete route survive a probe at 47#?\",\r\n \"verdict\": \"The defect at 47# is ordinary and the 47# enumeration cannot move item 1d's TPC threshold whatever value it returns; the bounded-defect Fekete lemma is stated exactly and proved with every hypothesis except the candidate itself discharged, so the route reduces to one named inequality and is neither closed nor open beyond that.\"\r\n }\r\n }\r\n ],\r\n \"registry_verdict\": \"The defect at 47# is ordinary and the 47# enumeration cannot move item 1d's TPC threshold whatever value it returns; the bounded-defect Fekete lemma is stated exactly and proved with every hypothesis except the candidate itself discharged, so the route reduces to one named inequality and is neither closed nor open beyond that.\"\r\n },\r\n {\r\n \"row\": 44,\r\n \"id\": \"Q-g2-falls-decision-rule\",\r\n \"todo\": \"1d\",\r\n \"status\": \"PARTIAL\",\r\n \"status_ok\": true,\r\n \"verdict_ok\": true,\r\n \"verdict\": \"CURRENT\",\r\n \"notes\": [\r\n {\r\n \"path\": \"research/history/staging/phase1-T3prep-decision-rule.md\",\r\n \"sha256\": \"08ab2b2be95121ef00cd99a0455ccc633bbb0088e436683822b4faa6afc94170\",\r\n \"ledger\": {\r\n \"id\": \"Q-g2-falls-decision-rule\",\r\n \"status\": \"PARTIAL\",\r\n \"todo\": \"1d\",\r\n \"question\": \"Does G2(x#)/x^2 fall, and what pre-registered decision rule would settle it?\",\r\n \"verdict\": \"The instrument is calibrated and sharp, resolving the one-class control at 11.6 sigma on the same nine points where it returns 0.9 sigma on G2, and the honest band on the slope is +/-0.117 at nine terms, so the rule is registered; but the power analysis puts separation at x = 53 only if the falling law is x ln^2 x and at x = 151 if it is x ln^3 x, so phase 1's extra terms settle nothing and no reachable exact ladder does either.\"\r\n }\r\n }\r\n ],\r\n \"registry_verdict\": \"The instrument is calibrated and sharp, resolving the one-class control at 11.6 sigma on the same nine points where it returns 0.9 sigma on G2, and the honest band on the slope is +/-0.117 at nine terms, so the rule is registered; but the power analysis puts separation at x = 53 only if the falling law is x ln^2 x and at x = 151 if it is x ln^3 x, so phase 1's extra terms settle nothing and no reachable exact ladder does either.\"\r\n },\r\n {\r\n \"row\": 45,\r\n \"id\": \"Q-hsub-reductions\",\r\n \"todo\": \"1d\",\r\n \"status\": \"PARTIAL\",\r\n \"status_ok\": true,\r\n \"verdict_ok\": true,\r\n \"verdict\": \"CURRENT\",\r\n \"notes\": [\r\n {\r\n \"path\": \"research/history/staging/attack-hsub-01.md\",\r\n \"sha256\": \"03626860c502880baeb35c7b9da215c69310749eecc29257717b2806ffd6d8d4\",\r\n \"ledger\": {\r\n \"id\": \"Q-hsub-reductions\",\r\n \"status\": \"PARTIAL\",\r\n \"todo\": \"1d\",\r\n \"question\": \"Can (H-sub) be proven or refuted, and what does the fold machinery reach?\",\r\n \"verdict\": \"Neither, but the statement is smaller: three reductions of the hypothesis are proven, the lemma consuming only power pairs so that it weakens to (H-sub-pow) with the conclusion unchanged, the fold machinery's inability to reach any of them is made exact, and the hunt over all 111 reachable pairs found no drifting family and no counterexample.\"\r\n }\r\n }\r\n ],\r\n \"registry_verdict\": \"Neither, but the statement is smaller: three reductions of the hypothesis are proven, the lemma consuming only power pairs so that it weakens to (H-sub-pow) with the conclusion unchanged, the fold machinery's inability to reach any of them is made exact, and the hunt over all 111 reachable pairs found no drifting family and no counterexample.\"\r\n },\r\n {\r\n \"row\": 46,\r\n \"id\": \"Q-xchannel-at23\",\r\n \"todo\": \"X\",\r\n \"status\": \"PARTIAL\",\r\n \"status_ok\": true,\r\n \"verdict_ok\": true,\r\n \"verdict\": \"CURRENT\",\r\n \"notes\": [\r\n {\r\n \"path\": \"research/history/staging/xchannel-at23.md\",\r\n \"sha256\": \"eaa77c372bdfe928858a8a00144aa45ec71e760fd5e2542eb3e47d67b35fe086\",\r\n \"ledger\": {\r\n \"id\": \"Q-xchannel-at23\",\r\n \"status\": \"PARTIAL\",\r\n \"todo\": \"X\",\r\n \"question\": \"What is the fifth point of the X-channel statistic, at @23, and does the monotone rise hold?\",\r\n \"verdict\": \"The fifth point is +0.2658 at @23, the rise holds at five levels and is larger than the four-point trend predicted, and the deficit has migrated into m >= 3, whose share of the X-gap runs 10.5, 23.5 and 81.1 per cent at @17, @19 and @23; the constant itself is not derived here.\"\r\n }\r\n }\r\n ],\r\n \"registry_verdict\": \"The fifth point is +0.2658 at @23, the rise holds at five levels and is larger than the four-point trend predicted, and the deficit has migrated into m >= 3, whose share of the X-gap runs 10.5, 23.5 and 81.1 per cent at @17, @19 and @23; the constant itself is not derived here.\"\r\n },\r\n {\r\n \"row\": 47,\r\n \"id\": \"Q-xchannel-closedform\",\r\n \"todo\": \"X\",\r\n \"status\": \"PARTIAL\",\r\n \"status_ok\": true,\r\n \"verdict_ok\": true,\r\n \"verdict\": \"CURRENT\",\r\n \"notes\": [\r\n {\r\n \"path\": \"research/history/staging/xchan-at29.md\",\r\n \"sha256\": \"e9ea5a647477176be2851a5c0162079e52a56db8a226d1c9225b900033e1fc64\",\r\n \"ledger\": {\r\n \"id\": \"Q-xchannel-closedform\",\r\n \"status\": \"PARTIAL\",\r\n \"todo\": \"X\",\r\n \"question\": \"Does the zero-parameter closed form for the joint deficit survive its two blind tests at @29 and @31?\",\r\n \"verdict\": \"1 - J = 4S2 is the only one of the three pre-registered laws left standing and it is not exact: a clean hit at @29, z = -0.90, and 4.93 sigma low at @31 where sigma_J falls to 0.000024, with the two rivals dying at 6.35 and 20.81 sigma; both blind levels agree on a stable relative offset of about half a percent, -0.41% and -0.48%, whose absolute residuals match to 2%.\"\r\n }\r\n }\r\n ],\r\n \"registry_verdict\": \"1 - J = 4S2 is the only one of the three pre-registered laws left standing and it is not exact: a clean hit at @29, z = -0.90, and 4.93 sigma low at @31 where sigma_J falls to 0.000024, with the two rivals dying at 6.35 and 20.81 sigma; both blind levels agree on a stable relative offset of about half a percent, -0.41% and -0.48%, whose absolute residuals match to 2%.\"\r\n },\r\n {\r\n \"row\": 48,\r\n \"id\": \"Q-xchannel-offset\",\r\n \"todo\": \"X\",\r\n \"status\": \"PARTIAL\",\r\n \"status_ok\": true,\r\n \"verdict_ok\": true,\r\n \"verdict\": \"CURRENT\",\r\n \"notes\": [\r\n {\r\n \"path\": \"research/history/staging/item-x-offset.md\",\r\n \"sha256\": \"56b36842f16025a152ab819929ff445179bb8a7b09ee063bf059198b25cfe867\",\r\n \"ledger\": {\r\n \"id\": \"Q-xchannel-offset\",\r\n \"status\": \"PARTIAL\",\r\n \"todo\": \"X\",\r\n \"question\": \"What produces the ~3.8 law's -0.45% offset, and which candidate correction survives?\",\r\n \"verdict\": \"The best available description is a finite-level correction of S3's size and sign - 1 - J = 4S2 - S3 scores z = +0.04 at @29 and -0.41 at @31 where the standing law scores -0.43 and -2.32 - but it is POST HOC; two candidate corrections and one shape are refuted on existing data, the sigma behind the @31 detection is calibrated on seven known-truth controls, and the deciding blind test at @37 is pre-registered and committed alone.\"\r\n }\r\n }\r\n ],\r\n \"registry_verdict\": \"The best available description is a finite-level correction of S3's size and sign - 1 - J = 4S2 - S3 scores z = +0.04 at @29 and -0.41 at @31 where the standing law scores -0.43 and -2.32 - but it is POST HOC; two candidate corrections and one shape are refuted on existing data, the sigma behind the @31 detection is calibrated on seven known-truth controls, and the deciding blind test at @37 is pre-registered and committed alone.\"\r\n },\r\n {\r\n \"row\": 49,\r\n \"id\": \"Q-at43-bigint-0830\",\r\n \"todo\": \"1\",\r\n \"status\": \"PARTIAL\",\r\n \"status_ok\": true,\r\n \"verdict_ok\": true,\r\n \"verdict\": \"CURRENT\",\r\n \"notes\": [\r\n {\r\n \"path\": \"research/history/staging/engine-0830-at43-bigint.md\",\r\n \"sha256\": \"fe2fff95ebc754eb6ec5ee326998c3e42c434f76fd0e9bc0b4b68e17974dac78\",\r\n \"ledger\": {\r\n \"id\": \"Q-at43-bigint-0830\",\r\n \"status\": \"PARTIAL\",\r\n \"todo\": \"1\",\r\n \"question\": \"Can the K-30 natal march be carried past its 2^53 ceiling to @43 exactly, and is the @43 point worth its cost?\",\r\n \"verdict\": \"The engine side is done and VERIFIED (five paths promoted to BigInt, @7..@37 and a 0.248% slice of @41 reproduced digit for digit); the @43 point is NOT run, because the measured extrapolation is 579 h of eight cores on this machine (a 43x tile times a 2.23x per-cell cost, against the brief's ~13x), which is a weeks-class box job whose only payoff is a sixth point on a curve with no consumer; the forecast 0.8393 raw / 0.8399 persisted stands unscored.\"\r\n }\r\n }\r\n ],\r\n \"registry_verdict\": \"The engine side is done and VERIFIED (five paths promoted to BigInt, @7..@37 and a 0.248% slice of @41 reproduced digit for digit); the @43 point is NOT run, because the measured extrapolation is 579 h of eight cores on this machine (a 43x tile times a 2.23x per-cell cost, against the brief's ~13x), which is a weeks-class box job whose only payoff is a sixth point on a curve with no consumer; the forecast 0.8393 raw / 0.8399 persisted stands unscored.\"\r\n },\r\n {\r\n \"row\": 50,\r\n \"id\": \"Q-comb-discrepancy\",\r\n \"todo\": \"8\",\r\n \"status\": \"PARTIAL\",\r\n \"status_ok\": true,\r\n \"verdict_ok\": true,\r\n \"verdict\": \"CURRENT\",\r\n \"notes\": [\r\n {\r\n \"path\": \"research/history/staging/comb-discrepancy-tight.md\",\r\n \"sha256\": \"b8464183b7543cfc83e0eee5832e5efa9cb07959c66e383efa174a28c2767871\",\r\n \"ledger\": {\r\n \"id\": \"Q-comb-discrepancy\",\r\n \"status\": \"PARTIAL\",\r\n \"todo\": \"8\",\r\n \"question\": \"Can the Comb Discrepancy Lemma's constant 2*3^k be tightened (TODO 8c)?\",\r\n \"verdict\": \"The lemma IS the trivial per-block bound; the attained optimum is max G - min G, a computed number per level (slack 29x @23, 54x @29, growing); term-by-term pricing lifts the certified head share to about 28-29 percent, not the 50 hoped for.\"\r\n }\r\n }\r\n ],\r\n \"registry_verdict\": \"The lemma IS the trivial per-block bound; the attained optimum is max G - min G, a computed number per level (slack 29x @23, 54x @29, growing); term-by-term pricing lifts the certified head share to about 28-29 percent, not the 50 hoped for.\"\r\n },\r\n {\r\n \"row\": 51,\r\n \"id\": \"Q-comb-tail-0830\",\r\n \"todo\": \"8\",\r\n \"status\": \"PARTIAL\",\r\n \"status_ok\": true,\r\n \"verdict_ok\": true,\r\n \"verdict\": \"CURRENT\",\r\n \"notes\": [\r\n {\r\n \"path\": \"research/history/staging/attack-0830-comb-tail.md\",\r\n \"sha256\": \"e40b8251f329407ad94677a24fdc9e8f231488e9a3e90b1a4d9914f8c892d40f\",\r\n \"ledger\": {\r\n \"id\": \"Q-comb-tail-0830\",\r\n \"status\": \"PARTIAL\",\r\n \"todo\": \"8\",\r\n \"question\": \"Can TODO item 8's remaining part (b), prime-comb equidistribution in the tail regime, be proven at the depths the certificate needs, or at least measured and priced at the enumerable levels (8b)?\",\r\n \"verdict\": \"Not proven and no exponent moves: the tail defect is measured exactly for the first time (signed -0.01% of the tail main term at @23, -0.00% at @29, unsigned 0.60% and 0.25%, smaller than the head's 1.00% in aggregate but 8 to 37% per prime at its worst), the Comb Discrepancy Lemma's method fails in the tail on its MAIN TERM (off by a derived factor e^gamma(1/u - q^(1-u)), measured aggregate 1.14x at @23) before its error term (10^27 x main), the prime-comb Legendre form has 2^(k+2) terms of which the short ones obey the lemma's <1 bound and the long ones are prime-in-AP discrepancies whose term-by-term sum is 11.2% of main at @23 against a signed -0.01%; the all-x form at K = 0 is Theorem C (PROVEN, limit only) and at the certificate's depth K* the non-closing step is a kappa = 1 asymptotic at sifting parameter s < 2, a proof gap on data that show B_tail = 0.998-1.000 at K* through @29; the full-depth half is 8(a), CLOSED.\"\r\n }\r\n }\r\n ],\r\n \"registry_verdict\": \"Not proven and no exponent moves: the tail defect is measured exactly for the first time (signed -0.01% of the tail main term at @23, -0.00% at @29, unsigned 0.60% and 0.25%, smaller than the head's 1.00% in aggregate but 8 to 37% per prime at its worst), the Comb Discrepancy Lemma's method fails in the tail on its MAIN TERM (off by a derived factor e^gamma(1/u - q^(1-u)), measured aggregate 1.14x at @23) before its error term (10^27 x main), the prime-comb Legendre form has 2^(k+2) terms of which the short ones obey the lemma's <1 bound and the long ones are prime-in-AP discrepancies whose term-by-term sum is 11.2% of main at @23 against a signed -0.01%; the all-x form at K = 0 is Theorem C (PROVEN, limit only) and at the certificate's depth K* the non-closing step is a kappa = 1 asymptotic at sifting parameter s < 2, a proof gap on data that show B_tail = 0.998-1.000 at K* through @29; the full-depth half is 8(a), CLOSED.\"\r\n },\r\n {\r\n \"row\": 52,\r\n \"id\": \"Q-c2prime-drift\",\r\n \"todo\": \"1c (retired)\",\r\n \"status\": \"PARTIAL\",\r\n \"status_ok\": true,\r\n \"verdict_ok\": true,\r\n \"verdict\": \"CURRENT\",\r\n \"notes\": [\r\n {\r\n \"path\": \"research/history/staging/attack-c2drift-01.md\",\r\n \"sha256\": \"2462c9e26796b6bd054298f4c355abe185fb3242d3b7ab9831cf1ae042843bf8\",\r\n \"ledger\": {\r\n \"id\": \"Q-c2prime-drift\",\r\n \"status\": \"PARTIAL\",\r\n \"todo\": \"1c (retired)\",\r\n \"question\": \"Why does c2' drift on the 22-term ladder?\",\r\n \"verdict\": \"The twin-specific mechanism for the drift is closed (cited as such by c2prime-refit-22.md).\"\r\n }\r\n },\r\n {\r\n \"path\": \"research/history/staging/c2prime-refit-22.md\",\r\n \"sha256\": \"1ed07f6258acee6cc233187e162e07397ed60e65b864e3417b48913503d70c56\",\r\n \"ledger\": {\r\n \"id\": \"Q-c2prime-drift\",\r\n \"status\": \"PARTIAL\",\r\n \"todo\": \"1c (retired)\",\r\n \"question\": \"Why does c2' drift on the 22-term ladder?\",\r\n \"verdict\": \"Class-count-blind (attack-c2drift-01 §1); on item 1c's own band x >= 41 the ten-point rate is unresolvable: c2' 0.3745 +/- 0.1411 against c1's 0.3344 +/- 0.2358 on the same primes, a difference of 0.0401 = 0.12 of the control's ten-point window sd. Whether the drift is real is NOT closed here: attack-c2drift-01 measured it at p = 0.0053 on x >= 17 and left (i)' and (ii) alive. exponent-control §2/§4/§6 refit at 22 terms.\"\r\n }\r\n }\r\n ],\r\n \"registry_verdict\": \"Class-count-blind (attack-c2drift-01 §1); on item 1c's own band x >= 41 the ten-point rate is unresolvable: c2' 0.3745 +/- 0.1411 against c1's 0.3344 +/- 0.2358 on the same primes, a difference of 0.0401 = 0.12 of the control's ten-point window sd. Whether the drift is real is NOT closed here: attack-c2drift-01 measured it at p = 0.0053 on x >= 17 and left (i)' and (ii) alive. exponent-control §2/§4/§6 refit at 22 terms.\"\r\n }\r\n]", "registry-check-38-52.out": "served research/QUESTIONS.md sha256(LF) e2ddcfc55f89be535041515a26fbc5068db832a7db548d35765b75f5360ff60e\r\nserved /questions: 217 questions, 53 open or partial\r\nsweeping rows 38..52: 15 rows, note set below\r\n\r\nROW 38 Q-rec-cheapest-0830 PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\r\nROW 39 Q-recon-0904-sifting-limit-floor PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\r\nROW 40 Q-rho-maximal-law PARTIAL status_ok=True verdict_ok=True -> CURRENT (2/2 note(s) with a ledger block)\r\nROW 41 Q-rho-sup-z41-0830 PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\r\nROW 42 Q-rml-proof-0829n PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\r\nROW 43 Q-fekete-1d-defect47 PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\r\nROW 44 Q-g2-falls-decision-rule PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\r\nROW 45 Q-hsub-reductions PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\r\nROW 46 Q-xchannel-at23 PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\r\nROW 47 Q-xchannel-closedform PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\r\nROW 48 Q-xchannel-offset PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\r\nROW 49 Q-at43-bigint-0830 PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\r\nROW 50 Q-comb-discrepancy PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\r\nROW 51 Q-comb-tail-0830 PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\r\nROW 52 Q-c2prime-drift PARTIAL status_ok=True verdict_ok=True -> CURRENT (2/2 note(s) with a ledger block)\r\n\r\n15 rows swept, 0 stale at the registry level.\r\nSilent on whether a ledger block is itself stale against its note body,\r\nriders and the outcome register: see rows23-37.md for that column.\r\n", "audit-report.md": "# Audit: `research/QUESTIONS.md` — four rows whose served source states the correction\n\n**Caveat first.** No mathematics and no status changes. All four rows stay PARTIAL; twin-prime\ninfinitude remains OPEN; no exponent moves. This audit changes **verdict text only**, and only\nwhere a served document already states the replacement verbatim, so that no new claim is\nintroduced. It is the second half of job #609 (the sweep report is the first).\n\n## Scope\n\n`GET /projects/twin-primes/questions` served 217 questions, 53 open or partial; the sweep covers\nrows 38–52 of that list (see `rows38-52.md`). All fifteen rows are faithful copies of their own\nledger blocks (REG 15/15 current), but **five are stale in their sources**. This audit corrects\nthe four whose served source supplies the repaired wording:\n\n| row | id | what changes |\n|---|---|---|\n| 38 | `Q-rec-cheapest-0830` | the ledger's comparator \"251 against z^{u0} = 1.05e6\" and the 16s/9 threshold, per RIDER 2026-08-30 |\n| 39 | `Q-recon-0904-sifting-limit-floor` | the \"METHOD ARTEFACT … about 18 per cent … passed, 0 of 15\" headline, per RIDER 2026-09-04 |\n| 47 | `Q-xchannel-closedform` | the unqualified \"@31 = 4.93 sigma\" reading, per the accepted audit #85's `also_fix` |\n| 48 | `Q-xchannel-offset` | \"the deciding blind test at @37 is pre-registered and committed alone\", per `Q-xchan-at37-score` (ANSWERED) |\n\n**Row 40 `Q-rho-maximal-law` is deliberately NOT patched.** No text in it is wrong: the defect is\nthat the generator's \"last note in path order\" rule displays the older of two same-id notes\n(`rho-maximal-law.md`, 2026-08-18/19/20) and hides the later `rho-exact-z31-01.md` (2026-08-28).\nA hand edit to that row would be reverted by the next regeneration; the fix belongs in\n`research/qc/questions.js` (pick the newest note, or emit `MIXED` when same-id verdicts disagree),\nwhich is what #149 proposed and no one has implemented. Reported here, not patched.\n\n## The patch\n\n`questions-rows38-52.patch` — a unified diff against `research/QUESTIONS.md`, **8 changed lines**\n(four verdicts × the two places each row appears: section 1 and section 2). Every replacement is\nthe superseding source's own wording:\n\n* **Row 38.** Keeps the ledger's structure and inserts the rider's two facts: the growth\n hypothesis **s ≤ 3**, and that the former comparator **1.05e6 is HM(u₀), not z^{u₀} = 7.188e7\n at z = 73**, so the gap is 68.3× larger than the ledger said. The 8.86e17 / 1.2e6 figures are\n untouched (not refuted).\n* **Row 39.** Records the rider's ruling as the rider states it — \"REFUTED as stated … on four\n grounds\" — and carries its two corrections: **22.3 per cent**, not 18, and the refused external\n test (a legal profile at x = 13 reads **5.0113**, above β₂, firing the note's own\n pre-registered refutation). The band sentence is unchanged.\n* **Row 47.** Applies accepted audit #85's `also_fix`: the verdict now says the @31 figure is\n quoted at the run's own σ_J and that the note's §7 calls it inherited, records the calibrated\n reading **z = −2.32** (inside the pre-registration's own HIT band, so \"one detection at 4.93\n sigma\" is zero detections), and names the adjacent row that quotes the other pair.\n* **Row 48.** Replaces the pre-@37 clause with the @37 outcome: `Q-xchan-at37-score` (ANSWERED,\n 2026-08-21) measured 1 − J = 0.020823, all seven candidates died at |z| = 104–122 with an EMPTY\n survivor set and a 13.4·S₃ shortfall, and the finite-level-correction family is CLOSED as the\n offset's description — this row's own 4S₂ − S₃ candidate with it. The @37 note's proposed\n item-X update is quoted as \"report only, not applied\".\n\n**It applies cleanly and is exactly what a regeneration produces for these rows** (the corrected\nverdicts are the notes' text, so a regenerated index agrees with the patch). Verified:\n\n```\nserved research/QUESTIONS.md sha256(LF) e2ddcfc55f89be535041515a26fbc5068db832a7db548d35765b75f5360ff60e\npatched research/QUESTIONS.md sha256(LF) e32ce3725dab552458d7df919f1f2e74a09bc883a8bd4463209c2a8bc18c35b8\n```\n\n`patch --binary -p1` on a copy reproduces the patched file byte for byte\n(`QUESTIONS.corrected.md`).\n\n## What must change downstream, or this patch does not hold\n\nThe patch fixes the **generated index**, not the **sources**. Four of the five stale rows are\nstale in their notes, so a regeneration after any unrelated ledger edit restores the old text\nunless the notes are corrected too. In the pattern #132 used for row 42 — the note's ledger plus a\ndated status rider — the source corrections are:\n\n* `attack-0830-rec-cheapest.md` — ledger clause per its own RIDER 2026-08-30 (row 38).\n* `recon-0904-sifting-limit-floor.md` — ledger headline per its own RIDER 2026-09-04 (row 39).\n* `xchan-at29.md` — ledger verdict per accepted audit #85's `also_fix` (row 47).\n* `item-x-offset.md` — final clause per `xchan-at37-score.md` §5, whose proposed update is\n already written there (row 48).\n\nI have not drafted those four note patches: #85's `also_fix` is already the drafted wording for\nrow 47, and #132 (accepted, pending integration) is the working pattern for the rest.\n\n## Falsifiers\n\n* **Row 38** is corrected wrongly if the 2026-08-30 rider's \"68.3× larger\" is itself wrong, i.e.\n if some other note defines the ledger's 1.05e6 as z^{u₀} rather than HM(u₀). The rider names\n `redteam-0830-floor-sign.md` as its source; I read the rider, not that note.\n* **Row 39** is corrected wrongly if the \"0 of 15\" external test did pass and the rider misread\n it. #121 (recorded) independently refuses the 18 % figure and leaves the test unscored; it does\n not settle the rider's stronger claim.\n* **Row 47** is corrected wrongly if the calibrated slot-clustered σ is the wrong model — which is\n the open question the correction makes explicit rather than hides.\n* **Row 48** is corrected wrongly if `xchan-at37-score.md`'s @37 census is later refuted; it is\n ANSWERED and independently recounted (`research/xchan-at37-02-verify.js`), so the clause is\n stale under any reading of its body.\n\n## Sources\n\n* `research/QUESTIONS.md` served, sha256 (LF) `e2ddcfc55f89…`; corrected sha256 (LF)\n `e32ce3725dab…`. `research/history/staging/attack-0830-rec-cheapest.md` (RIDERs 2026-08-30,\n 2026-09-04); `recon-0904-sifting-limit-floor.md` (RIDER 2026-09-04); `xchan-at29.md`;\n `item-x-offset.md`; `xchan-at37-score.md` sha256 `3252f9611a27ce27…` §5.\n* Returns cited: #54, #84, #85 (review #34), #121, #130, #131, #132, #149, #205.\n\n## Files\n\n| file | what it is |\n|---|---|\n| `questions-rows38-52.patch` | the four-row correction, LF, `-p1` |\n| `QUESTIONS.corrected.md` | the applied result (sha256(LF) `e32ce372…`) |\n| `mkpatch3852.py` | builds it; asserts each old string occurs exactly twice and the line count is preserved |\n| `registry-check.py`, `registry-check-38-52.out` | the REG test that shows the index is faithful — i.e. that the fix belongs upstream |\n| `rows38-52.md` | the fifteen-row table with all citations |\n", "questions-rows38-52.patch": "--- a/research/QUESTIONS.md\n+++ b/research/QUESTIONS.md\n@@ -173,5 +173,5 @@\n | 0 | `Q-measure-g2-generic` Where does G2(x#) sit in the distribution of the same maximum gap taken over other two-class configurations with the same census, and is the constant-shift family generic inside the free-pair family? | ANSWERED | G2 is GENERIC in this coordinate on the registered rule, MEASURED: the midrank percentile of G2 in the census-matched two-class ensemble reads 20.00, 18.06, 26.35, 17.14 exhaustively at x = 11..19 and 13.70 sampled at x = 23, all inside the registered [5, 95] band so the falsifier does not fire, while sitting about one ensemble sd below the mean at all five levels; and the free-pair and constant-shift ensembles are PROVEN identical up to translation, so killer 1 gains no target in the class positions at these levels. | [measure-g2-generic-0829.md](history/staging/measure-g2-generic-0829.md) |\n | 0 | `Q-omega-floor-blind-0830` Does the local log-slope of the pointwise floor Omega(z) of attack-0830-rec-cheapest.md (exact to z = 73, certified 2+4-chain family to z = 5e5), extended blind to new levels under a sealed forecast, behave as the DERIVED growth law Omega >> z^{16s/9}/ln^8 z predicts at s = 2.698721, and at what z would the test be decisive? | ANSWERED | MEASURED, blind, on the derived law only: 18 of 22 sealed rows HIT; the certified 2+4-chain family at p* = (47, 43) runs 5e5 -> 1e9 (A1A2 = 8.863e17 -> 9.351e32, exact integers) with its step slope falling 4.85 -> 4.38 through every sealed band onto the law's log-corrected slope, the decisive decade 1e8 -> 1e9 reading 4.410 against 16s/9 - 8/ln z = 4.389 (HIT, kill rule not triggered at 4.452 +/- 0.011); the 4 MISSes are the ratio to z^{16s/9}/ln^8 z sitting at 2.00-2.14 above the sealed cap of 2 at z >= 1e8, the pre-declared non-adverse direction; the exact floor is extended 73 -> 113 (251 -> 1980, equal to the hill-climb floors at 89 and 101), where the law predicts nothing; no exponent moves and the law stays DERIVED and HELD. | [blind-0830-omega-floor.md](history/staging/blind-0830-omega-floor.md) |\n-| 0 | `Q-rec-cheapest-0830` At the cheapest legal point of REC(s, u0), s -> 1+sqrt(e) and u0 -> beta_2, where the saving asked over the triangle inequality is only z^{1.031}, does any of the five dead routes close, does a mixed mean-square-plus-cap strategy close, and is the failure a proof gap or a truth gap? | PARTIAL | No route closes and the failing step is yesterday's statement, so PARTIAL on the routes; but the step is located as a TRUTH gap, not a proof gap, asymptotically: the certificate's own value at a CRT-planted doubly-smooth window is cc(r) = -(A1A2 + A1B2 + B1A2) with every term a count of exit chains of the Rosser supports (PROVEN, checked exhaustively at z <= 73), the window bound T <= cc(r) + H - 1 is exact, and the exit-chain count is DERIVED (held) to grow like z^{16s/9 - o(1)} at four primes and z^{2s - o(1)} in the limit, so REC(s, u0) is false for every u0 < 16s/9 = 4.7088 at s -> 1+sqrt(e), the whole legal band (2, beta_2] included, and RML(alpha) with it; at every computable z it is a proof gap (Omega(z) exact to z = 73 reads 251 against z^{u0} = 1.05e6, and a certified lower bound 8.86e17 at z = 5e5 sits a factor 1.2e6 under z^{u0}); no exponent moved. | [attack-0830-rec-cheapest.md](history/staging/attack-0830-rec-cheapest.md) |\n+| 0 | `Q-rec-cheapest-0830` At the cheapest legal point of REC(s, u0), s -> 1+sqrt(e) and u0 -> beta_2, where the saving asked over the triangle inequality is only z^{1.031}, does any of the five dead routes close, does a mixed mean-square-plus-cap strategy close, and is the failure a proof gap or a truth gap? | PARTIAL | No route closes and the failing step is yesterday's statement, so PARTIAL on the routes; but the step is located as a TRUTH gap, not a proof gap, asymptotically: the certificate's own value at a CRT-planted doubly-smooth window is cc(r) = -(A1A2 + A1B2 + B1A2) with every term a count of exit chains of the Rosser supports (PROVEN, checked exhaustively at z <= 73), the window bound T <= cc(r) + H - 1 is exact, and the exit-chain count is DERIVED (held) to grow like z^{16s/9 - o(1)} for s <= 3, the hypothesis the 2026-08-30 red team added to the threshold, and z^{2s - o(1)} in the limit, so REC(s, u0) is false for every u0 < 16s/9 = 4.7088 at s -> 1+sqrt(e), the whole legal band (2, beta_2] included, and RML(alpha) with it, its falsity beyond z ~ 5.6e31; at every computable z it is a proof gap (Omega(z) exact to z = 73 reads 251, and the ledger's former comparator 1.05e6 is HM(u0), not z^{u0} = 7.188e7 at z = 73, so the gap is 68.3x larger than the ledger stated, while a certified lower bound 8.86e17 at z = 5e5 sits a factor 1.2e6 under z^{u0}); no exponent moved. | [attack-0830-rec-cheapest.md](history/staging/attack-0830-rec-cheapest.md) |\n | 0 | `Q-recon-0828-covering` Does the interval-covering, automated-proof or extremal-combinatorics literature hold a method that bounds the length of an interval covered by two residue classes per prime, or that turns the exact G₂ ladder into a statement at all x? | ANSWERED | No route. Fifteen angles examined, fourteen dead on arrival or already closed here; the best interval-covering bound in print is Crittenden-Vanden Eynden's own Lemma 2, which reads 1.02e12 against G₂(79#) = 1710 and is 8181x weaker than the corpus's own 4.2665 sieve bound at the same level; the one survivor, transplanting Costello-Watts's recurrent certificate to two classes, is a computation that certifies finitely many levels and whose analytic half is Erdős #970 at one class. | [recon-0828-covering.md](history/staging/recon-0828-covering.md) |\n | 0 | `Q-recon-0828-farfields` Do any of the fields still absent from IMPORT-MAP.md hold a theorem whose hypothesis the tile satisfies exactly and whose conclusion is a lower bound on a periodic set's count in every window, a maximal-gap bound for a product-structured periodic set, or a positivity statement at a distinguished point? | ANSWERED | No route from any of the sixteen candidates priced; two reach STRONG-ANALOGY or better and CLEAN and both bank only a closure or a wall address, the Fourier one by a one-line non-vanishing proof that empties the Donoho-Stark/Tao/Meshulam family on this object, and Janson's lower tail by saturating at exp(-1/c) >= 0.0805 against the 1/W a position union bound needs, short by 3.6e2 at x = 11 and 2.8e13 at x = 41. | [recon-0828-farfields.md](history/staging/recon-0828-farfields.md) |\n@@ -181,5 +181,5 @@\n | 0 | `Q-recon-0829-farfields2` Do any of eight fields still absent from IMPORT-MAP.md and from the three 2026-08-28 recon notes hold a theorem whose hypothesis the tile or the zone satisfies exactly and whose conclusion is a lower bound on a periodic set's count in every window, a maximal-gap bound for a product-structured periodic set, or a positivity statement at a distinguished point? | ANSWERED | No route from any of the eight; the pass banks one PUBLISHED-ANCHOR (Holt and Rudd's transfer matrix has its full spectrum in print, second eigenvalue a2 = prod_{17<=q<=p}(q-3)/(q-2), printed 0.10206751799779 at p = 999999999989, decaying only like 2.82/ln p, and this corpus's prior-art record names the eigenstructure and the binomial eigenvectors but carries the per-prime eigenvalue (p-j-1)/(p-2) at lit-pdf-holt-rudd.md line 259 but no product-over-primes closed form, no numerical value and no rate), one CLOSURE with mechanism (Beurling-Selberg is the proof device behind the large sieve, whose constant N-1+1/delta is an l2-to-l2 statement and whose error 1/delta exceeds the window at the operative theta_total > 2, so the whole extremal-function family reproduces a row already rejected), and three wall addresses; two angles are TPC-strength at the form that would pay and are labelled (ii). | [recon-0829-farfields2.md](history/staging/recon-0829-farfields2.md) |\n | 0 | `Q-recon-0830-rec-killrun` Does the literature hold, in its own conventions, a theorem whose hypotheses REC(s, u0) (the sup-versus-rms recovery of the level-D signed remainder over all positions, attack-0829n-rml-proof.md section 3) or the weighted kill-run K*(s) (the longest run of level-s slots the primes in (s, 2s] can kill, attack-0829n-doubling-bridge.md sections 0 and 3) satisfies as stated, or a Maier-type theorem that makes REC false for two-class sifted sets at some u0 below beta_2? | ANSWERED | No theorem applies to either object as stated, on four calibrated channels searched in the owning conventions; every neighbour is graded NEAREST with its unmet hypothesis named, and neither object is on any refuted row. On the Maier question the answer is negative for REC as stated (the Maier family bounds the COUNT, i.e. the full-level remainder, never a level-D truncation) but carries one calibration: at one class and full level the REC-shaped inequality is FALSE asymptotically, by Buchstab's origin ratio against the Montgomery-Vaughan full-period variance ceiling, and the embedded arithmetic puts the level where that falsity first shows at log10 z between 16 and 141 depending on u0 and epsilon, so a finite-z margin of the kind the corpus measures cannot see a failure of this type; at two classes the same full-level statement is HL-conditional at u0 = 2 and unlocated in print at any u0. No exponent moved. | [recon-0830-rec-killrun.md](history/staging/recon-0830-rec-killrun.md) |\n-| 0 | `Q-recon-0904-sifting-limit-floor` What is actually known, in the owning convention and at the page, about LOWER bounds on sifting limits at dimension kappa > 1, and is the class-blind cap at beta_2 = 4.26645 a barrier or a method artefact? | PARTIAL | Both. The corpus's [ABSENT] is wrong for lower bounds and right for extremal examples: Selberg's Lectures section 17 and Brady's 2017 Corollaries 1 and 3 ARE lower bounds on beta(kappa) at kappa > 1, invisible because Selberg's convention is the reciprocal a_k = 1/beta_kappa so his lower bounds are titled \"upper bounds for sifting limits\", while no kappa > 1 extremal example exists and Halberstam's own 2003 review says so; the best floor on beta(2) resting on a citation is 1.8394 from Brady's reduction evaluated at kappa = 2, unconditional and below the band, and beta(2) >= 2 follows from monotonicity by a padding argument DERIVED here and probably folklore; the cap at 4.26645 is a METHOD ARTEFACT at rung MEASURED, since the level-D LP is exactly the axiom-only information class and its calibrated floor of 3.3152 sits about 18 per cent below the DHR value and passed, 0 of 15, its first external test against the published beta_3 and beta_4; the band (2, 4.26645] is untouched by anything proven. | [recon-0904-sifting-limit-floor.md](history/staging/recon-0904-sifting-limit-floor.md) |\n+| 0 | `Q-recon-0904-sifting-limit-floor` What is actually known, in the owning convention and at the page, about LOWER bounds on sifting limits at dimension kappa > 1, and is the class-blind cap at beta_2 = 4.26645 a barrier or a method artefact? | PARTIAL | Both. The corpus's [ABSENT] is wrong for lower bounds and right for extremal examples: Selberg's Lectures section 17 and Brady's 2017 Corollaries 1 and 3 ARE lower bounds on beta(kappa) at kappa > 1, invisible because Selberg's convention is the reciprocal a_k = 1/beta_kappa so his lower bounds are titled \"upper bounds for sifting limits\", while no kappa > 1 extremal example exists and Halberstam's own 2003 review says so; the best floor on beta(2) resting on a citation is 1.8394 from Brady's reduction evaluated at kappa = 2, unconditional and below the band, and beta(2) >= 2 follows from monotonicity by a padding argument DERIVED here and probably folklore; the cap at 4.26645 as a METHOD ARTEFACT at rung MEASURED is REFUTED as stated (rider 2026-09-04, from `redteam-0904-sifting-limit.md`, on four independent grounds): the level-D LP fixes one omega-profile out of the class Ford defines by a one-sided inequality, and a legal profile inside the classical budget at x = 13 reads 5.0113 calibrated, above beta_2, firing the note's own pre-registered refutation; the ledger's \"3.3152 sits about 18 per cent below the DHR value\" is 22.3 per cent; and its \"passed, 0 of 15\" external test does not survive, so the calibrated sentence is that no barrier theorem exists, no axiom-only argument inside the band is exhibited, and the LP number is a finite-level measurement for one profile; the band (2, 4.26645] is untouched by anything proven. | [recon-0904-sifting-limit-floor.md](history/staging/recon-0904-sifting-limit-floor.md) |\n | 0 | `Q-redteam-0828-closures` Do the five 2026-08-28 closure notes stand, and do they stand for the reasons they give? | ANSWERED | Three stand on corrected sentences (1d, Z5b, 0); one stands in half (10, the 1.074 counter-claim does not); one does not stand for its stated reason (1c, closed on a narrower window than the prior work's significant one, and its CLOSED has overwritten a PARTIAL in the generated index). | [redteam-0828-closures.md](history/staging/redteam-0828-closures.md) |\n | 0 | `Q-redteam-0830-floor-growth` Does the DERIVED growth half of attack-0830-rec-cheapest.md sec.4.2 (Omega >> z^{16s/9}/ln^8 z from four-prime Rosser exit chains) survive an adversarial re-derivation on independent code, does it really make REC(s, u0) false for every u0 < 16s/9, and does the blind slope test of blind-0830-omega-floor.md test the LAW or the CONSTRUCTION? | ANSWERED | No break found in six attacks, so the growth half SURVIVES this pass and stays DERIVED (one failed break is not a proof, and the whole thing still rests on the exact half E1-E3 that this pass did not grade); but four claims are corrected - the threshold 16s/9 needs the unstated hypothesis s <= 3 (above it the four-prime LP maximum is 2(2s+2)/3, then 8, both below 16s/9, though the corollary that every legal u0 < beta_2 fails survives at every admissible s because 2 x LP-max > beta_2 for all s > 9beta_2/16 = 2.3999), the 2s - o(1) limit needs 3^k <= ln D / ln 8 so only k = 2 is available at every z below 1.1e9, A_i is >= not ~ the dyadic count, and the certified ratio does NOT rise monotonically (0.0118, 0.0128, 0.0101, 0.0057, 0.0051, 0.0093, 0.0336, 0.0801, 0.1658, 0.3545 on my own recount, which reproduces all ten S4b rows digit for digit); the blind test is ruled a test of the CONSTRUCTION's count and not of the law, since it cannot see E1-E3 or the CRT step; on the ruling Chris asked for, the floor does make the RML route a truth gap asymptotically and conditionally on E1-E4, and changes nothing computable, the literal construction first passing H = z^{u0} at z ~ 5.6e31 on the constant 2.93e-4 measured here; no exponent moved. | [redteam-0830-floor-growth.md](history/staging/redteam-0830-floor-growth.md) |\n@@ -219,6 +219,6 @@\n | X | `Q-xchan-at37-score` What does the @37 census say about the sealed X-channel offset pre-registration? | ANSWERED | The census measured 1 - J = 0.020823, below every registered prediction, so scored exactly as registered every one of the seven candidates dies at \\|z\\| = 104 to 122 and the survivor set is EMPTY, an outcome the sealed prereg has no consequence clause for; the number itself survives an independent recount on a different marking scheme, and the offset question is replaced by a larger one. | [xchan-at37-score.md](history/staging/xchan-at37-score.md) |\n | X | `Q-xchannel-at23` What is the fifth point of the X-channel statistic, at @23, and does the monotone rise hold? | PARTIAL | The fifth point is +0.2658 at @23, the rise holds at five levels and is larger than the four-point trend predicted, and the deficit has migrated into m >= 3, whose share of the X-gap runs 10.5, 23.5 and 81.1 per cent at @17, @19 and @23; the constant itself is not derived here. | [xchannel-at23.md](history/staging/xchannel-at23.md) |\n-| X | `Q-xchannel-closedform` Does the zero-parameter closed form for the joint deficit survive its two blind tests at @29 and @31? | PARTIAL | 1 - J = 4S2 is the only one of the three pre-registered laws left standing and it is not exact: a clean hit at @29, z = -0.90, and 4.93 sigma low at @31 where sigma_J falls to 0.000024, with the two rivals dying at 6.35 and 20.81 sigma; both blind levels agree on a stable relative offset of about half a percent, -0.41% and -0.48%, whose absolute residuals match to 2%. | [xchan-at29.md](history/staging/xchan-at29.md) |\n-| X | `Q-xchannel-offset` What produces the ~3.8 law's -0.45% offset, and which candidate correction survives? | PARTIAL | The best available description is a finite-level correction of S3's size and sign - 1 - J = 4S2 - S3 scores z = +0.04 at @29 and -0.41 at @31 where the standing law scores -0.43 and -2.32 - but it is POST HOC; two candidate corrections and one shape are refuted on existing data, the sigma behind the @31 detection is calibrated on seven known-truth controls, and the deciding blind test at @37 is pre-registered and committed alone. | [item-x-offset.md](history/staging/item-x-offset.md) |\n+| X | `Q-xchannel-closedform` Does the zero-parameter closed form for the joint deficit survive its two blind tests at @29 and @31? | PARTIAL | 1 - J = 4S2 is the only one of the three pre-registered laws left standing and it is not exact: a clean hit at @29, z = -0.90, and 4.93 sigma low at @31 where sigma_J falls to 0.000024, with the two rivals dying at 6.35 and 20.81 sigma; both blind levels agree on a stable relative offset of about half a percent, -0.41% and -0.48%, whose absolute residuals match to 2%. The @31 figure is quoted at the run's own sigma_J = 0.000024, and audit #85 (accepted, verified, its also_fix unapplied) records that the note's section 7 calls that sigma inherited rather than derived: under the slot-clustered sigma calibrated in `item-x-offset.md` section 4 on seven known-truth controls the same residual reads z = -2.32, inside the pre-registration's own HIT band, so this row's \"one detection at 4.93 sigma\" is zero detections at the pre-registration's threshold; the adjacent `Q-xchannel-offset` row quotes -0.43 and -2.32 for this same law at the same two levels, a factor 2.09 and 2.13 apart. | [xchan-at29.md](history/staging/xchan-at29.md) |\n+| X | `Q-xchannel-offset` What produces the ~3.8 law's -0.45% offset, and which candidate correction survives? | PARTIAL | The best available description is a finite-level correction of S3's size and sign - 1 - J = 4S2 - S3 scores z = +0.04 at @29 and -0.41 at @31 where the standing law scores -0.43 and -2.32 - but it is POST HOC; two candidate corrections and one shape are refuted on existing data, the sigma behind the @31 detection is calibrated on seven known-truth controls, and the deciding blind test at @37 has RUN: `Q-xchan-at37-score` (ANSWERED, 2026-08-21) measures 1 - J = 0.020823 there, below every registered prediction, so scored exactly as registered every one of the seven candidates dies at |z| = 104 to 122 with a 13.4*S3 shortfall and an EMPTY survivor set, and the finite-level-correction family is CLOSED as the offset's description, this row's 4S2 - S3 candidate with it; the @37 note's proposed TODO item X update is recorded there as \"report only, not applied\". | [item-x-offset.md](history/staging/item-x-offset.md) |\n | X | `Q-xchannel-triples` Does the joint deficit deepen or decay as the mixed super-W triple census is extended past @17? | ANSWERED | It decays: J = 0.9025, 0.9599, 0.9659 at @17, @19, @23, so the deficit runs 9.7%, 4.0%, 3.4%, both new points land inside the pre-registered DECAYING FASTER THAN FORCED band, and the DEEPENING instinct is refuted by 3.50 sigma between @17 and @19 and a further 1.68 sigma to @23. | [xchannel-triples.md](history/staging/xchannel-triples.md) |\n | A | `Q-anchored-ladder` Does the unified anchored ladder generalize past @11 and @13, and what does cheap K buy at each level? | ANSWERED | It generalizes verbatim to @17 and @19, and the K = 0 unified floor is DEAD there (vacuous at -1262 and -55865); what survives and grows is the family lead at the classic certification depth K*, margin +2, +5, +26, +110 across @11 to @19. | [attack-anchored-02.md](history/staging/attack-anchored-02.md) |\n@@ -647,5 +647,5 @@\n | `Q-reachability-coverage` | ANSWERED | Which part of the original divisor domain can no improvement to the grouped moment's nonzero kernel reach, in either orientation, and what would a uniform saving on that kernel actually buy? | For a prescribed fixed exponent margin eta0, the zero budgets leave S_0={p<2eta0,q<2eta0} below that required margin regardless of the nonzero-kernel saving. Uniform gamma=2 covers its complement at that prescribed margin. Smaller positive margins can reach fixed interior points of S_0; only p=q=0 remains at zero budget one for every margin. The proposed eta0~loglog x/log x substitution is not justified by estimates carrying x^epsilon losses; its box count is geometry only. The sufficient twin margin remains OPEN. Independent review F3 and F5 correct the quantifiers and the prime-power tail proof; round-review-0906 corrects the diagonal scope, fixed-margin bulk claim and six-cut separation. | C | [reachability-coverage.md](reachability-coverage.md) |\n | `Q-readings-traceability` | ANSWERED | What is the source of each of the 1,592 figures that appear in a script's READINGS and not in its own OUTPUT block? | The running record: nothing in the pass changes a number, 56 mismatches are classified in eight kinds for adjudication elsewhere, and the advisory count rises when the work is done right because a provenance declaration is prose that carries printed values. | none | [readings-traceability.md](history/staging/readings-traceability.md) |\n-| `Q-rec-cheapest-0830` | PARTIAL | At the cheapest legal point of REC(s, u0), s -> 1+sqrt(e) and u0 -> beta_2, where the saving asked over the triangle inequality is only z^{1.031}, does any of the five dead routes close, does a mixed mean-square-plus-cap strategy close, and is the failure a proof gap or a truth gap? | No route closes and the failing step is yesterday's statement, so PARTIAL on the routes; but the step is located as a TRUTH gap, not a proof gap, asymptotically: the certificate's own value at a CRT-planted doubly-smooth window is cc(r) = -(A1A2 + A1B2 + B1A2) with every term a count of exit chains of the Rosser supports (PROVEN, checked exhaustively at z <= 73), the window bound T <= cc(r) + H - 1 is exact, and the exit-chain count is DERIVED (held) to grow like z^{16s/9 - o(1)} at four primes and z^{2s - o(1)} in the limit, so REC(s, u0) is false for every u0 < 16s/9 = 4.7088 at s -> 1+sqrt(e), the whole legal band (2, beta_2] included, and RML(alpha) with it; at every computable z it is a proof gap (Omega(z) exact to z = 73 reads 251 against z^{u0} = 1.05e6, and a certified lower bound 8.86e17 at z = 5e5 sits a factor 1.2e6 under z^{u0}); no exponent moved. | 0 | [attack-0830-rec-cheapest.md](history/staging/attack-0830-rec-cheapest.md) |\n+| `Q-rec-cheapest-0830` | PARTIAL | At the cheapest legal point of REC(s, u0), s -> 1+sqrt(e) and u0 -> beta_2, where the saving asked over the triangle inequality is only z^{1.031}, does any of the five dead routes close, does a mixed mean-square-plus-cap strategy close, and is the failure a proof gap or a truth gap? | No route closes and the failing step is yesterday's statement, so PARTIAL on the routes; but the step is located as a TRUTH gap, not a proof gap, asymptotically: the certificate's own value at a CRT-planted doubly-smooth window is cc(r) = -(A1A2 + A1B2 + B1A2) with every term a count of exit chains of the Rosser supports (PROVEN, checked exhaustively at z <= 73), the window bound T <= cc(r) + H - 1 is exact, and the exit-chain count is DERIVED (held) to grow like z^{16s/9 - o(1)} for s <= 3, the hypothesis the 2026-08-30 red team added to the threshold, and z^{2s - o(1)} in the limit, so REC(s, u0) is false for every u0 < 16s/9 = 4.7088 at s -> 1+sqrt(e), the whole legal band (2, beta_2] included, and RML(alpha) with it, its falsity beyond z ~ 5.6e31; at every computable z it is a proof gap (Omega(z) exact to z = 73 reads 251, and the ledger's former comparator 1.05e6 is HM(u0), not z^{u0} = 7.188e7 at z = 73, so the gap is 68.3x larger than the ledger stated, while a certified lower bound 8.86e17 at z = 5e5 sits a factor 1.2e6 under z^{u0}); no exponent moved. | 0 | [attack-0830-rec-cheapest.md](history/staging/attack-0830-rec-cheapest.md) |\n | `Q-recon-0828-covering` | ANSWERED | Does the interval-covering, automated-proof or extremal-combinatorics literature hold a method that bounds the length of an interval covered by two residue classes per prime, or that turns the exact G₂ ladder into a statement at all x? | No route. Fifteen angles examined, fourteen dead on arrival or already closed here; the best interval-covering bound in print is Crittenden-Vanden Eynden's own Lemma 2, which reads 1.02e12 against G₂(79#) = 1710 and is 8181x weaker than the corpus's own 4.2665 sieve bound at the same level; the one survivor, transplanting Costello-Watts's recurrent certificate to two classes, is a computation that certifies finitely many levels and whose analytic half is Erdős #970 at one class. | 0 | [recon-0828-covering.md](history/staging/recon-0828-covering.md) |\n | `Q-recon-0828-farfields` | ANSWERED | Do any of the fields still absent from IMPORT-MAP.md hold a theorem whose hypothesis the tile satisfies exactly and whose conclusion is a lower bound on a periodic set's count in every window, a maximal-gap bound for a product-structured periodic set, or a positivity statement at a distinguished point? | No route from any of the sixteen candidates priced; two reach STRONG-ANALOGY or better and CLEAN and both bank only a closure or a wall address, the Fourier one by a one-line non-vanishing proof that empties the Donoho-Stark/Tao/Meshulam family on this object, and Janson's lower tail by saturating at exp(-1/c) >= 0.0805 against the 1/W a position union bound needs, short by 3.6e2 at x = 11 and 2.8e13 at x = 41. | 0 | [recon-0828-farfields.md](history/staging/recon-0828-farfields.md) |\n@@ -657,5 +657,5 @@\n | `Q-recon-0830-rec-killrun` | ANSWERED | Does the literature hold, in its own conventions, a theorem whose hypotheses REC(s, u0) (the sup-versus-rms recovery of the level-D signed remainder over all positions, attack-0829n-rml-proof.md section 3) or the weighted kill-run K*(s) (the longest run of level-s slots the primes in (s, 2s] can kill, attack-0829n-doubling-bridge.md sections 0 and 3) satisfies as stated, or a Maier-type theorem that makes REC false for two-class sifted sets at some u0 below beta_2? | No theorem applies to either object as stated, on four calibrated channels searched in the owning conventions; every neighbour is graded NEAREST with its unmet hypothesis named, and neither object is on any refuted row. On the Maier question the answer is negative for REC as stated (the Maier family bounds the COUNT, i.e. the full-level remainder, never a level-D truncation) but carries one calibration: at one class and full level the REC-shaped inequality is FALSE asymptotically, by Buchstab's origin ratio against the Montgomery-Vaughan full-period variance ceiling, and the embedded arithmetic puts the level where that falsity first shows at log10 z between 16 and 141 depending on u0 and epsilon, so a finite-z margin of the kind the corpus measures cannot see a failure of this type; at two classes the same full-level statement is HL-conditional at u0 = 2 and unlocated in print at any u0. No exponent moved. | 0, D | [recon-0830-rec-killrun.md](history/staging/recon-0830-rec-killrun.md) |\n | `Q-recon-0830-smooth-aps` | PARTIAL | Does any theorem on y-friable integers in arithmetic progressions (Granville, Fouvry-Tenenbaum, Soundararajan, Harper, Drappeau, Drappeau-Granville-Shao) supply the uniform o(1) equidistribution that attack-0830-varE-identification.md section 4 leaves open, for moduli up to y^(4/5) with the lam1 weight and the squarefree restriction? | NONE APPLIES AS STATED, and the step stays open; lim Var/E = 0.45546 stays HEURISTIC. Read at the page (eleven sources, arXiv text layers and the Acta Math. page image): every pointwise asymptotic is O(log q / log y)-precise (Granville 1993 Thm 1) or hypothesises log x / log q -> infinity, and every beyond-square-root level (Fouvry-Tenenbaum 1996, Drappeau 2015, Drappeau-Granville-Shao 2017, Pascadi 2023/2025) hypothesises y <= x^delta, i.e. u -> infinity, against our u in (1.2, 2]. NEAREST: Harper arXiv:1208.5992 (2012, a preprint with no journal version found) Theorem 2, the Barban-Davenport-Halberstam form, whose range (log^K x <= y <= x, Q <= Psi/log^A x, bounded u in the ineffective form) covers d up to L^(1/2)/log and whose weight (1 on all friable n, no squarefree restriction, no max over x' <= x) is unmet. The record's quantifier is also refined: the cell needs the lam0-weighted AVERAGE over d, not uniformity. PROVEN given the weighted form (H_w) of that theorem, derivation mine and unreviewed, with its two identities checked numerically in the producer: every two-branch cell of Xmix, both mirror cases, all bands, balanced blocks included, is o(ln^2 y), so the Bettin-Chandee kernel separation is not on this route; the reduction of (H_w) to Harper's theorem is OUTLINED and not written, and the three-branch type (measured -0.0002 at x = 19, trivial bound O(ln^3 y)) is untouched. Not TPC-strength. | 9 | [recon-0830-smooth-aps.md](history/staging/recon-0830-smooth-aps.md) |\n-| `Q-recon-0904-sifting-limit-floor` | PARTIAL | What is actually known, in the owning convention and at the page, about LOWER bounds on sifting limits at dimension kappa > 1, and is the class-blind cap at beta_2 = 4.26645 a barrier or a method artefact? | Both. The corpus's [ABSENT] is wrong for lower bounds and right for extremal examples: Selberg's Lectures section 17 and Brady's 2017 Corollaries 1 and 3 ARE lower bounds on beta(kappa) at kappa > 1, invisible because Selberg's convention is the reciprocal a_k = 1/beta_kappa so his lower bounds are titled \"upper bounds for sifting limits\", while no kappa > 1 extremal example exists and Halberstam's own 2003 review says so; the best floor on beta(2) resting on a citation is 1.8394 from Brady's reduction evaluated at kappa = 2, unconditional and below the band, and beta(2) >= 2 follows from monotonicity by a padding argument DERIVED here and probably folklore; the cap at 4.26645 is a METHOD ARTEFACT at rung MEASURED, since the level-D LP is exactly the axiom-only information class and its calibrated floor of 3.3152 sits about 18 per cent below the DHR value and passed, 0 of 15, its first external test against the published beta_3 and beta_4; the band (2, 4.26645] is untouched by anything proven. | 0 | [recon-0904-sifting-limit-floor.md](history/staging/recon-0904-sifting-limit-floor.md) |\n+| `Q-recon-0904-sifting-limit-floor` | PARTIAL | What is actually known, in the owning convention and at the page, about LOWER bounds on sifting limits at dimension kappa > 1, and is the class-blind cap at beta_2 = 4.26645 a barrier or a method artefact? | Both. The corpus's [ABSENT] is wrong for lower bounds and right for extremal examples: Selberg's Lectures section 17 and Brady's 2017 Corollaries 1 and 3 ARE lower bounds on beta(kappa) at kappa > 1, invisible because Selberg's convention is the reciprocal a_k = 1/beta_kappa so his lower bounds are titled \"upper bounds for sifting limits\", while no kappa > 1 extremal example exists and Halberstam's own 2003 review says so; the best floor on beta(2) resting on a citation is 1.8394 from Brady's reduction evaluated at kappa = 2, unconditional and below the band, and beta(2) >= 2 follows from monotonicity by a padding argument DERIVED here and probably folklore; the cap at 4.26645 as a METHOD ARTEFACT at rung MEASURED is REFUTED as stated (rider 2026-09-04, from `redteam-0904-sifting-limit.md`, on four independent grounds): the level-D LP fixes one omega-profile out of the class Ford defines by a one-sided inequality, and a legal profile inside the classical budget at x = 13 reads 5.0113 calibrated, above beta_2, firing the note's own pre-registered refutation; the ledger's \"3.3152 sits about 18 per cent below the DHR value\" is 22.3 per cent; and its \"passed, 0 of 15\" external test does not survive, so the calibrated sentence is that no barrier theorem exists, no axiom-only argument inside the band is exhibited, and the LP number is a finite-level measurement for one profile; the band (2, 4.26645] is untouched by anything proven. | 0 | [recon-0904-sifting-limit-floor.md](history/staging/recon-0904-sifting-limit-floor.md) |\n | `Q-reconcile-bind` | ANSWERED | Can the unbound script tails be reconciled against their code and then bound by embed.js? | Six producers came back CLEAN-BOUND and one REBOUND with two documented label supersessions and identical results; nothing was DEFERRED and nothing DRIFTED outside the one pre-registered band move, fdecay-band-01's point 35.5 to 33.9 and band [30.8, 42.9] to [29.6, 38.3]. | none | [reconcile-bind.md](history/staging/reconcile-bind.md) |\n | `Q-record-deficit` | MIXED (lit-kourbatov-shortfall.md: ANSWERED; record-location-null.md: PARTIAL) | Does the 6.0 percent record-location deficit survive the corrected null, and what is it? | Survives (z = -3.3, five window cuts); Z5's premises on 'location' and on n_eff were wrong; corrected-null formula in §4. | Z5 | [lit-kourbatov-shortfall.md](history/staging/lit-kourbatov-shortfall.md), [record-location-null.md](history/staging/record-location-null.md) |\n@@ -812,6 +812,6 @@\n | `Q-xchan-at37-score` | ANSWERED | What does the @37 census say about the sealed X-channel offset pre-registration? | The census measured 1 - J = 0.020823, below every registered prediction, so scored exactly as registered every one of the seven candidates dies at \\|z\\| = 104 to 122 and the survivor set is EMPTY, an outcome the sealed prereg has no consequence clause for; the number itself survives an independent recount on a different marking scheme, and the offset question is replaced by a larger one. | X | [xchan-at37-score.md](history/staging/xchan-at37-score.md) |\n | `Q-xchannel-at23` | PARTIAL | What is the fifth point of the X-channel statistic, at @23, and does the monotone rise hold? | The fifth point is +0.2658 at @23, the rise holds at five levels and is larger than the four-point trend predicted, and the deficit has migrated into m >= 3, whose share of the X-gap runs 10.5, 23.5 and 81.1 per cent at @17, @19 and @23; the constant itself is not derived here. | X | [xchannel-at23.md](history/staging/xchannel-at23.md) |\n-| `Q-xchannel-closedform` | PARTIAL | Does the zero-parameter closed form for the joint deficit survive its two blind tests at @29 and @31? | 1 - J = 4S2 is the only one of the three pre-registered laws left standing and it is not exact: a clean hit at @29, z = -0.90, and 4.93 sigma low at @31 where sigma_J falls to 0.000024, with the two rivals dying at 6.35 and 20.81 sigma; both blind levels agree on a stable relative offset of about half a percent, -0.41% and -0.48%, whose absolute residuals match to 2%. | X | [xchan-at29.md](history/staging/xchan-at29.md) |\n-| `Q-xchannel-offset` | PARTIAL | What produces the ~3.8 law's -0.45% offset, and which candidate correction survives? | The best available description is a finite-level correction of S3's size and sign - 1 - J = 4S2 - S3 scores z = +0.04 at @29 and -0.41 at @31 where the standing law scores -0.43 and -2.32 - but it is POST HOC; two candidate corrections and one shape are refuted on existing data, the sigma behind the @31 detection is calibrated on seven known-truth controls, and the deciding blind test at @37 is pre-registered and committed alone. | X | [item-x-offset.md](history/staging/item-x-offset.md) |\n+| `Q-xchannel-closedform` | PARTIAL | Does the zero-parameter closed form for the joint deficit survive its two blind tests at @29 and @31? | 1 - J = 4S2 is the only one of the three pre-registered laws left standing and it is not exact: a clean hit at @29, z = -0.90, and 4.93 sigma low at @31 where sigma_J falls to 0.000024, with the two rivals dying at 6.35 and 20.81 sigma; both blind levels agree on a stable relative offset of about half a percent, -0.41% and -0.48%, whose absolute residuals match to 2%. The @31 figure is quoted at the run's own sigma_J = 0.000024, and audit #85 (accepted, verified, its also_fix unapplied) records that the note's section 7 calls that sigma inherited rather than derived: under the slot-clustered sigma calibrated in `item-x-offset.md` section 4 on seven known-truth controls the same residual reads z = -2.32, inside the pre-registration's own HIT band, so this row's \"one detection at 4.93 sigma\" is zero detections at the pre-registration's threshold; the adjacent `Q-xchannel-offset` row quotes -0.43 and -2.32 for this same law at the same two levels, a factor 2.09 and 2.13 apart. | X | [xchan-at29.md](history/staging/xchan-at29.md) |\n+| `Q-xchannel-offset` | PARTIAL | What produces the ~3.8 law's -0.45% offset, and which candidate correction survives? | The best available description is a finite-level correction of S3's size and sign - 1 - J = 4S2 - S3 scores z = +0.04 at @29 and -0.41 at @31 where the standing law scores -0.43 and -2.32 - but it is POST HOC; two candidate corrections and one shape are refuted on existing data, the sigma behind the @31 detection is calibrated on seven known-truth controls, and the deciding blind test at @37 has RUN: `Q-xchan-at37-score` (ANSWERED, 2026-08-21) measures 1 - J = 0.020823 there, below every registered prediction, so scored exactly as registered every one of the seven candidates dies at |z| = 104 to 122 with a 13.4*S3 shortfall and an EMPTY survivor set, and the finite-level-correction family is CLOSED as the offset's description, this row's 4S2 - S3 candidate with it; the @37 note's proposed TODO item X update is recorded there as \"report only, not applied\". | X | [item-x-offset.md](history/staging/item-x-offset.md) |\n | `Q-xchannel-triples` | ANSWERED | Does the joint deficit deepen or decay as the mixed super-W triple census is extended past @17? | It decays: J = 0.9025, 0.9599, 0.9659 at @17, @19, @23, so the deficit runs 9.7%, 4.0%, 3.4%, both new points land inside the pre-registered DECAYING FASTER THAN FORCED band, and the DEEPENING instinct is refuted by 3.50 sigma between @17 and @19 and a further 1.68 sigma to @23. | X | [xchannel-triples.md](history/staging/xchannel-triples.md) |\n | `Q-y2-recompute` | ANSWERED | Does the sixteen-level Y2 ladder reproduce on the corrected instrument, and does Y2/x^2 fall? | The ladder reproduces exactly, sixteen of sixteen from sixteen independent processes, digit for digit against the published 2026-08-18 ladder, and Y2/x^2 falls at 15 of the 15 consecutive steps; this does NOT show that G2(x#)/x^2 falls, since the shortfall Y2/(G2-1) is unmeasured above x = 79. | 1d | [y2-recompute.md](history/staging/y2-recompute.md) |\n", "QUESTIONS.corrected.md": "# Questions: what has been asked here, what it got, and where the record is\n\n*(GENERATED from the `` blocks at the top of the notes; never\nhand-edited. Regenerate with `node research/qc.js --index` or\n`node research/gen-questions-index.js`. The block format and the gate that\nenforces it are documented at the top of `qc/questions.js`.)*\n\n**Before proposing an attack or briefing an agent:** find the TODO item in\nsection 1, search for the actual object, then read the relevant records and\ntheir decisive dependencies. Use [OUTCOMES.md](OUTCOMES.md) for reuse and\nrevisit conditions. An item can span many unrelated attempts; loading every\nlinked note is not the required first step. Verify any conclusion that\ndetermines the next move, including an ANSWERED or CLOSED entry.\n\n**Status describes the question, not proof strength or review grade.**\nOPEN: unresolved. PARTIAL: specified parts are resolved. ANSWERED: the bounded\nquestion has an answer, which may be a negative finding, finite measurement\nor completed specification. Its underlying arithmetic target may remain OPEN.\nCLOSED: the stated premise or approach is closed at its recorded scope; this\ndoes not refute every related conjecture. SUPERSEDED: another record replaces\nthis one. Consult the owning note and linked reviews for HELD or other grades.\n\n**Multiple records:** conflicting statuses are shown as MIXED. The displayed\nverdict comes from the last record in path order, not necessarily the newest\nor most authoritative review. Read the linked records before resolving a\ndisagreement; the generator does not adjudicate it.\n\n## 1. By TODO item: what has already run under each\n\n| TODO item | question | status | verdict | records |\n|---|---|---|---|---|\n| C | `Q-agent-start` What must an incoming agent read and preserve before continuing the reviewed arithmetic campaign? | ANSWERED | Onboarding and document ownership only. The second dispatch is completed and independently reviewed on 2026-09-09. RESEARCH-EXECUTION.md owns current dispositions and candidate next obligations; RESEARCH-HANDOFF.md owns the updated exact contract. No assignment is started by that review. Twin-prime infinitude and required signed estimates remain OPEN. | [AGENT-START.md](AGENT-START.md) |\n| C | `Q-bilinear-fold-first-attack` Does assigning each Mobius factor to its first fold, or applying a direct second-moment bound, establish the missing one-sided estimate? | ANSWERED | The first-fold identity yields no saving; a direct second moment has a negligible diagonal and an open two-prime off-diagonal. Fixed-fold covariance is an exact residue-imbalance square and the beta weights balance at each fixed modulus. Growing-depth joint control remains OPEN; B>=-C2*x+o(x) is not improved. | [bilinear-fold-attack.md](bilinear-fold-attack.md) |\n| C | `Q-centered-discrepancy-estimate` Can the top of D_y be removed using ordinary prime distribution after an exact divisor flip, and what estimate remains? | PARTIAL | Equations (3)-(4), the exact split and flip, survive review and finite controls. A repair of the top-range bound T^top=O_(A,eps)(x/log^A x) is DERIVED in section 3a (2026-09-08) with the endpoint atom retained, odd square divisors, the corrected reciprocal-totient local factors with (m',g)=1, explicit truncation powers, a Cauchy multiplicity device, and a uniform Mobius mean proved from the q=1 Siegel-Walfisz statements; its inputs are the terminal-point BV of Tao Notes 3 Theorem 17 with the prefix form derived by rounding. The handler (lane V) read the derivation on 2026-09-08, reconstructed (BV*), (3a.9), (3a.14) and (3a.16), checked the limiting constant of (3a.9) numerically at eight (b,g) pairs and found no defect; it is accepted at its stated scope, with ineffective constants. The payoff is D_y=D^(e_1)+O_(A,eps)(x/log^A x) only, so the handoff consumer is equivalent to D^(e_1)>=-4x/25+o(x). The fixed-endpoint signed discrepancy, D_y>=-4x/25+o(x) and twin-prime infinitude remain OPEN. | [centered-discrepancy-estimate.md](centered-discrepancy-estimate.md) |\n| C | `Q-centered-discrepancy-measurement` At reachable x, is the OPEN sufficient input D_y>=-4x/25 numerically violated, is the absolute form (13) numerically plausible, and does D_y carry structure beyond a random-sign model? | ANSWERED | MEASURED to x=2^38. Neither pre-registered falsifier fires. D_y/x stays within 0.0043 of zero for the measured j>=26. Over 30<=j<=38, D_y/x differs from -(T1/x-C2) by at most 2e-4, with maximum 1.84401e-4 at j=30; the classical term dominates this finite comparison. The data do not establish an asymptotic fluctuation scale, an impossibility of informative future measurements, or the sufficient signed estimate. No proof status changes. | [centered-discrepancy-measurement.md](centered-discrepancy-measurement.md) |\n| C | `Q-chain-review-0906` Do the written derivations from S(x) down to E_dagger hold as stated, with every imported theorem used inside its hypotheses and every discarded term at the claimed O_H(x/log^H x) rate, and where does the chain remain unreviewed? | ANSWERED | Three independent adversarial passes found no defect in the six classical contributions from S(x) to E_>, none in the grouped moment (1)-(20) including the E_* to E_dagger step, and none in the residual-coverage link E_> to E_*, whose block-shape derivation was reconstructed and verified exactly. Every imported theorem was confirmed in its primary source except the Mobius Bombieri-Vinogradov citation, which points at an exercise sheet and needs a published replacement. Stale o(x) rates in three note bodies were replaced. A checked argument is not a refereed theorem; the OPEN signed margin is untouched. | [chain-review-0906.md](chain-review-0906.md) |\n| C | `Q-chen-fold-benchmark` Can the fold framework reproduce a classical Chen lower bound on long intervals with every imported hypothesis and error budget stated? | ANSWERED | The classical implication is derived below using three explicitly imported theorem statements, with the combinatorial minorant proved and its integral margin certified by exact rational arithmetic. This is a benchmark, not a new prime theorem, and it does not prove twins or a short-zone result. | [chen-fold-benchmark.md](chen-fold-benchmark.md) |\n| C | `Q-chen-opportunity` Does the signed Chen target identify an approachable proof mechanism, and what should the next research obligation be? | ANSWERED | The target is sufficient but subtracts an avoidable odd-composite penalty; even cancellation of its obvious signed components does not close the existing separate bounds. Select a prime-detecting weight against explicitly justified arithmetic input ranges before pursuing a new estimate. The opportunity is a research judgment, not a proved mechanism. | [chen-opportunity-audit.md](chen-opportunity-audit.md) |\n| C | `Q-chen-signed-target` What exact additional arithmetic estimate would turn the completed Chen benchmark into a proof of infinitely many twin primes? | PARTIAL | An elementary prime minorant gives a conditional twin theorem with all weights, ranges, quantifiers and errors specified. The opportunity audit identifies an avoidable negative-weight penalty and a separate-estimate budget that fails even under component cancellation; this sufficient target remains OPEN and is not the sole next proof obligation. | [chen-signed-target.md](chen-signed-target.md) |\n| C | `Q-coefficient-structure` Does the arithmetic structure of the aggregated coefficients reduce the endpoint estimate beyond the previous rectangle, and can smaller coefficient norms alone supply the remaining power saving? | ANSWERED | Exact classification gives a nonsquarefree L2 norm of size at most sqrt(D)*W^(1/4) times logarithms, while the squarefree top-band energy is at least a constant times D*W*log(W). At d~x^0.277, e~x^0.467 every endpoint sector except the squarefree-squarefree prime sum has a power saving. This reduces that rectangle to one explicit endpoint correlation; prime-dispersion.md supplies the remaining estimate and controls the full rectangle. Finite algebra, CRT sectors and archived prefixes are checked separately. | [coefficient-structure.md](coefficient-structure.md) |\n| C | `Q-cofactor-progression-transfer` Can the corrected full-cofactor kernel be estimated on a growing cofactor range by representing its actual arithmetic on CRT progressions, and does the resulting bound reach the twin consumer? | PARTIAL | Derived from named imports: for some positive absolute kappa and d, all prime-r cofactor pairs s,t <= floor((log x)^kappa), including non-squarefree inputs and mixed transition/smoothed profiles, admit a dyadic scale-average O(log^(2-d) X) bound after division by x. The extension uses bounded multiplicative Euler modifications on the progression s\\|n, t\\|n-2 and keeps its density 1/lcm(s,t), so summing cofactors costs only O((log log X)^2). This goes beyond s=t=1 but remains short of o(x). The large-cofactor joint remainder, outside residual and twin margin remain OPEN; the attempted extension to power-sized cofactors fails on source modulus range and accumulated logarithmic cost. | [cofactor-progression-transfer.md](cofactor-progression-transfer.md) |\n| C | `Q-consumer-comparison` How does the sufficient margin C_2 x + E_dagger(x) >= c_0 x/(log x)^K compare logically with the published sufficient routes to twin primes, and what does the literature already own of the corner object? | ANSWERED | The sufficient margin remains OPEN. The published conditional routes imply it where their hypotheses and stated lower bounds apply; reverse implications are not established. The dictionary is E_dagger(x)/x = C_2(2 delta_x - delta_{x/2} - 1) + o(1), with K>0 below that normalisation's resolution. Independent review corrects the comparison: cumulative and dyadic consumers on unbounded scales are equivalent at the same fixed K with possibly different positive constants. At K=0 an eventual cumulative lower bound also implies the unbounded-scale dyadic consumer. The nonnegative-weight corner identification applies only to the prime-cofactor subfamily; the full residual retains additional branches. Every surveyed route remains conditional. | [consumer-comparison.md](consumer-comparison.md) |\n| C | `Q-corner-branch-diagnostic` Can future experiments enumerate the full coefficients C(n) and C'(n-2) branch by branch, without silently dropping proper prime powers, cofactors s>1, non-squarefree inputs or either gcd branch? | ANSWERED | The driver exists and its two independent constructions of each side agree exactly on 8192 retained sides and 1802 fixture sides; five deletion and reconstruction controls are ACTIVE in both runs. Its split by gcd(n,n-2) measures integer parity, not the per-term CRT gcd partition. The fixed-eta corner is empty on the retained prefix, and Elo exceeds E0 across the entire J_x, so the matrix is a proxy window. This is finite algebra and tool preparation, with no asymptotic estimate or twin margin. | [corner-branch-diagnostic.md](corner-branch-diagnostic.md) |\n| C | `Q-corner-coefficient-energy` Can the full corner coefficients be studied before splitting their cofactor branches, and what do exact energy and classical smoothed sieve weights supply? | ANSWERED | The sharp coefficient energy is an explicit signed quadratic form in short cofactors plus O_epsilon(x^(1/2)K^2*x^epsilon), a power below x at the fixed corner. Logarithmically averaging the divisor cutoff gives Barban–Vehov weights and an O_eta(x log x) squared norm on each side and absolute shifted-product bound. The follow-up sharp-corner-transition.md gives sharp O_eta(x log^2 x) norms and proves a non-negligible norm transition for sufficiently small fixed eta. Signed transition saving and the global complement remain OPEN; no exact residual cut or twin margin changes. | [corner-coefficient-energy.md](corner-coefficient-energy.md) |\n| C | `Q-corner-correlation` On the corner S_0 where both cofactors sit just above their own cutoffs, what exactly is the remaining endpoint sum, what does the sufficient consumer need there, and does any reviewed correlation theorem supply it? | PARTIAL | The full corner is exactly sum_n C(n)C'(n-2). Proper-prime-power terms above the cutoffs are negligible at fixed eta by §1.1; the remaining prime-r terms with s>1 or s'>1 and non-squarefree inputs remain OPEN. The s=s'=1 piece has a nonnegative Mobius-product weight with exact moving cuts. The multiplicative lift and round-review sampling give small continuous and dyadic scale-average log savings for that subfamily, with no o(x), full-corner control or twin margin. The complement is still open. | [corner-correlation.md](corner-correlation.md) |\n| C | `Q-corner-log-average` What do logarithmically averaged correlation theorems and their quantitative successors actually supply for the prime-cofactor corner weight, with exact support, rates and scale quantifiers? | PARTIAL | The pure band is invariant under dilation by primes outside it; the exact cofactor window is not, but its support extends to x^(1-2eta) and need not leave the window under a fixed dilation. Uniform-prefix logarithmic bounds imply block bounds by Abel summation; a relative exponent above 3 is one sufficient route to o(x), not a universal necessary threshold. Pilatte's 1/(96e) calculation applies only to his displayed parameter choice. A bounded multiplicative Fourier lift now gives a small continuous scale-average log saving for the prime-cofactor subfamily via Tao-Teravainen v2; see prime-band-transfer.md. No o(x), dyadic pointwise estimate, full-corner estimate or positive twin margin follows. | [corner-log-average.md](corner-log-average.md) |\n| C | `Q-corner-measurement` At finite dyadic x, how large is the corner two-point correlation K(x)=sum_{x/2(x)>=c*x/log^K x on unbounded dyadic scales suffices, as does a stated logarithmically rescaled average. These implications are derived from named inputs; their endpoint hypotheses remain OPEN. The recorded uniform-gap theorem comparison has different quantifiers and does not establish an obstruction for this consumer. | [endpoint-target-audit.md](endpoint-target-audit.md) |\n| C | `Q-fixed-endpoint-discrepancy` After the accepted truncation to odd moduli e=-4x/25+o(x)? | PARTIAL | Reviewed 2026-09-09 after the coprimality repair: T_I^low=O_(A,eps')(x/log^A x), hence S=C_2x+B+O_A(x/log^A x), with B the exact Type II plus band sum (2.9). The untruncated modulus bound was false; g<=(log x)^(A+13), both tails, the coprime density and the weighted BV multiplicity now pay the estimate. Review corrects harmless log powers and the power-of-two atom over the full eps' range. D^(e_1)>=-4x/25+o(x) is equivalent to B+2C_2M>=-4x/25+o(x); it implies, but is not equivalent to, B>=-(C_2-1/200)x+o(x). The latter and H_B are sufficient OPEN margins. The absolute band statement (4.9) is one stronger sufficient input, not a necessary condition. No inspected source estimates the remaining actual signed B. Twin-prime infinitude remains OPEN. | [fixed-endpoint-discrepancy.md](fixed-endpoint-discrepancy.md) |\n| C | `Q-fold-arithmetic-bridge` Does one parity-table bridge from the anchored fold ledger yield a sufficient twin lower bound with named arithmetic inputs? | PARTIAL | Exact identities retained. The one unread sieve input of the pricing, Bombieri--Vinogradov for k-fold X^(1/u)-rough products, is derived from Wu's Lemma 2.3 (section 3a); the same input, sieved in the composite variable, gives the contamination aggregate constant 4 for each fixed k,u, replacing the displayed 12.86 to 19.72. With these inputs, elementary bounds give Q_cov(u)<1 and c*_real(u)<4 for every u>4 (section 4a, independently reviewed 2026-09-09 with rational certificates), so neither sufficient ratio test succeeds at any depth. This closes the two tests, not the decorrelation hypotheses, and supplies no twin estimate. | [fold-arithmetic-bridge.md](fold-arithmetic-bridge.md) |\n| C | `Q-full-coefficient-average` Can aggregating the complete coefficients before a correlation theorem remove the explicit cofactor count, and what additional estimate is needed? | PARTIAL | Exact factor and rounded-endpoint Fourier identities retained, with c_(i,0)=3/5 and an explicit composite-filtered weighted sum. The full family is not 1-bounded, but the sufficient phase condition admits at least k=0,+/-1 on the left and l=0,+/-1,...,+/-6 on the right for every Mellin twist; the earlier zero-only claim is corrected. Composite filtering and the required correlation rate remain unmatched. Proposition 6.5, independently reviewed including on 2026-09-09, proves coefficient norm at least (log x)^(2/5) for representations by 1-bounded functions on all smooth inputs. This does not exclude density-one representations, paid growing components or a jointly treated Fourier sum. No sufficient signed twin margin follows. | [full-coefficient-average.md](full-coefficient-average.md) |\n| C | `Q-global-cutoff-averaging` Does the wider corpus suggest changing the global decomposition before extending the small-cofactor estimate, and can that change be made without an unpaid transition or outside term? | PARTIAL | Derived: independently averaging the two initial Mobius cutoffs over fixed exponent intervals gives S=C2*x+sum G_L(n)G_R(n-2)+O_A(x/log^A x), uniformly on dyadic scales, with all cofactor and prime-power branches included. Each full G has squared norm O(x log x). The existing sharp-corner lower norm then forces cancellation between the corner coefficient and the rest at the same input. This does not estimate their shifted product. The global signed residual was already O(x) by the sieve upper bound and positivity; the new norm representation is not an improved signed bound. Prefer a bounded attempt on this global coefficient pair and a one-sided consumer; the twin margin remains OPEN. | [global-cutoff-averaging.md](global-cutoff-averaging.md) |\n| C | `Q-global-factor-signs` Which full factor configurations cause negative global products, can prime-power exceptions be paid, and does a pair-trigger upper bound control the negative part? | PARTIAL | Derived: G_i(n)=Lambda_(>W_i)(n)-F_i(s_i(n))*log t_i(n)-E_i(n), with all small-prime factors in s_i and all primes of t_i exceeding W_i. The correction is supported on a small-prime power exceeding W_i; its full shifted effect is O_epsilon(x^(39/40+epsilon)). After also paying proper prime powers, the residual is a composite-filtered signed smooth/rough cofactor sum. A ten-small-prime cell has F=-84 although every pair lies below a, refuting the proposed pair-trigger majorant. Prime filters contribute explicit terms when bounding the negative part; dropping them is an upper bound, not an identity. No scale-x signed improvement or twin margin follows. | [global-factor-signs.md](global-factor-signs.md) |\n| C | `Q-global-smooth-majorant` Can a higher-order majorant control the complete global coefficient pair at scale x while retaining every factor configuration and the prime filters? | PARTIAL | Derived using the corrected Henriot upper theorem: a C3 probability average of the same admissible initial cutoffs gives a full residual with sum \\|Ghat_L(n) Ghat_R(n-2)\\|=O(x). The three-smallest-prime majorant has bounded harmonic mass and meets the source growth class uniformly, despite not being multiplicative. All prime-power exceptions are paid. This replaces the O(x log x) absolute budget for the earlier logarithmic profile by O(x) for a different admissible profile representing the same signed residual to arbitrary logarithmic precision. The implied constant is not compared with C2 and no improved signed lower bound or twin margin is supplied. | [global-smooth-majorant.md](global-smooth-majorant.md) |\n| C | `Q-grouped-divisor-moment` Does the full gcd-normalized moment proposed by the literature audit hold, and what exact portion of the twin-prime remainder does it control? | ANSWERED | The proposed moment is derived from classical completion with all coefficient sectors and uniform twists included. Full rectangles are controlled when delta<19/25 and delta+3nu<161/100, in addition to the preceding region. A concrete extra cut d<=floor(x^(151/200)), de^3<=floor(x^(321/200)) controls the entire d~e~x^(2/5) benchmark and leaves an explicit smaller-domain endpoint remainder. The uniform product threshold stays below 19/25. At delta=8/25,nu=9/20, the next deficit is confined to small-common-divisor nonzero kernels; their required saving and the global twin margin remain OPEN. Finite validation does not prove asymptotic rates. | [grouped-divisor-moment.md](grouped-divisor-moment.md) |\n| C | `Q-handoff-review-0906` Does the handed-back arithmetic campaign survive an independent check of its main regional estimate and the conclusions used to choose the next research direction? | ANSWERED | The bounded handoff audit is completed in reports 20 and 21: the checked local reduction and regional mechanisms survive, named source statements were verified, and the consumer, corner support, rate, shrinking-margin and identity-piece overclaims were corrected. Joint Cauchy is now priced and adds no region. The multiplicative band transfer has a separate PARTIAL owner with a weaker continuous-scale payoff. Imported deep theorems remain imports; this is not corpus-wide certification or a twin margin. | [handoff-review-0906.md](handoff-review-0906.md) |\n| C | `Q-heath-brown-edges` Does the a=1 edge, and hence the corner S_0, come from the particular two-cutoff Vaughan identity in prime-detection-spec.md and shifted-prime-decomposition.md, or does an equivalent edge reappear under Heath-Brown's identity and the other standard decompositions of Lambda? | ANSWERED | The exact identities and exponent simplex leave a=1 in separately bounded formal pieces, including Heath-Brown's balanced j=K configuration. This does not prove an intrinsic edge after identity-piece cancellation. The bounded-cofactor object has affine dilations, not literally F(n)F(n-2). Linnik's fixed truncation has a large term-wise absolute tail by an elementary counting proof; its signed tail is not thereby bounded below. Heath-Brown K=3 narrows the classical split band, with coefficient and Type I normalization obligations retained. No full alternative reduction, new region or twin margin follows. | [heath-brown-edges.md](heath-brown-edges.md) |\n| C | `Q-joint-correction-source-audit` Does completing the supported prime-partner correction yield cancellation beyond the existing Vaughan reduction, and which literature inputs actually justify that completion? | ANSWERED | CHECKED prior-art matches for subset convolution, complementary sieve sums and signed smooth-number distribution. DERIVED reconstruction identifies the unrestricted factor weight with an existing short divisor approximant, up to the paid prime-power error. Its supports are x^(12/25) and x^(1/10). Reassembling the entire joint correction returns the already-open Vaughan remainder and supplies no new signed bound. A formal factor-vector control distinguishes the unrestricted logarithmic weight from the selected prime-cofactor branch. The matched general sieve theorems require inputs not supplied here. Literature novelty is unestablished; twin infinitude remains OPEN. | [joint-correction-source-audit.md](joint-correction-source-audit.md) |\n| C | `Q-joint-factor-estimate` Can an additional estimate for actual factor configurations control the full signed balance of the C3 residual? | ANSWERED | DERIVED: for the single-large-prime delta-rough family F, the residual splits exactly as R_F = C2*x*H_delta - Z_F^p + o(x), where the Type I part C2*x*H_delta is evaluated by ordinary prime BV on each cofactor interval and the twisted Mobius constant 2C2 (corrected 2026-09-08 after V2: Wu Lemma 2.3 is squarefree-only and did not cover the non-squarefree b_R moduli; prime BV covers all moduli), and Z_F^p is the signed prime-partner sum. The complement carries the opposite Type I part -C2*x*H_delta, so grouping by factor configuration transfers Type I mass and leaves the twisted prime correlation B_L unchanged. With the matched upper sieve the family bound is C2*x(1+H_delta-(40/9)H^+), certified below -2/5 C2*x: INSUFFICIENT BOUND. Even a sieve constant 2 in place of 40/9 needs H^abs<1 and leaves the complement unpaid. Independently reviewed 2026-09-09 with the uniform delta-range and complement corrections below. Twin infinitude and every sufficient margin remain OPEN. | [joint-factor-estimate.md](joint-factor-estimate.md) |\n| C | `Q-kernel-sign-control` At delta=8/25, nu=9/20, is the moment Mfrak and its small-common-divisor kernel X_small measurably smaller with the actual coefficient b_u=A_right(gu) than with \\|b_u\\| or with \\|b_u\\| times an independent seeded random sign, and does the actual/random ratio decrease with x over the feasible dyadic range? | ANSWERED | MEASURED, negative. Over seven dyadic scales j=18..30 and four boxes, the pre-registered slope of log2(actual/random) for X_small has mean + sd >= 0 in all four families, so the run gives no heuristic support for extra Mobius-sign cancellation in the small-j kernel. The actual coefficient's \\|X_small\\| ranks like one more random draw. The measurement is weak where it matters: the finite moment is 97.66-102.58 percent equal-frequency class, X_small carries 0.02-2.88 percent, J0=floor(x^(1/20)) is 1 at j=18 and 2 at j=20..30, and Z=2 collapses the prime-power sector of (13) to r=2. Equation (21) and the global twin margin remain OPEN and untouched. | [kernel-sign-control.md](kernel-sign-control.md) |\n| C | `Q-left-divisor-signs` Can the Möbius structure of the left coefficient A_left(gm) replace the first Cauchy inequality, so that its sqrt(M) loss is not paid, and what does that buy at the target box (delta,nu)=(8/25,9/20) and at the corner a=b=1? | PARTIAL | Partly, and not where it is needed. Two lemmas are derived: a Type I bound with no zero-frequency budget, usable at the target box whenever the arbitrary block is shorter than x^(1/4), and a Type II bound that is uniform in the Perron twist heights. Applied to the actual coefficient they confine the target-box deficit to sectors with left prime power r>=x^(43/200) and right prime power q>x^(3/100), and lower the worst block exponent there from 103/100 to 41/40. The corner a=b=1 is untouched: grouped, Type I, Type II, Bettin-Chandee/Wright and even a hypothetical per-block square-root bound all exceed 1 there. No region is added, E_dagger and the consumer (20) are unchanged, and the global twin margin remains OPEN. | [left-divisor-signs.md](left-divisor-signs.md) |\n| C | `Q-mobius-bv-derivation` Is there a published theorem statement of Bombieri-Vinogradov for the Mobius function at level T^(1/2)/log^B T with a maximum over coprime classes, and if not, does that estimate follow from published theorems by a derivation whose every hypothesis is checked? | ANSWERED | No published theorem statement was located in five channels searched in the owning convention named in research/SEARCH-CONVENTIONS.md; the closest is Granville-Shao's published assertion that the result is known, which carries no proof locator for the Mobius case. The estimate is derived here from four numbered results of Koukoulopoulos, GSM 203 (Corollary 13.4, Theorem 26.2, Theorem 26.6, equation (26.3)) plus Vaughan's identity for mu, with B(A)=A+6 and an ineffective constant. This is a derivation from published theorems, not a refereed theorem, and it is read from the author's preliminary version of that book. | [mobius-bv-derivation.md](mobius-bv-derivation.md) |\n| C | `Q-moving-cutoff-parity` Does a proof-level reading of the shifted-Mobius sieve give a valid, explicit error-tolerant arithmetic target, and does its divisor switching preserve the moving boundary? | ANSWERED | CHECKED prior art: Murty--Vatwani already supplies the conditional squarefree/parity mechanism and fixed-residue reduction. Its printed p. 654 swap omits n+h>ey; an exact finite counterexample verifies failure of that equality. DERIVED dyadic repair keeps the boundary and gives S=C2*x-2*C2*M+D_y+O_A(x/log^A x). A one-sided D_y>=-4*x/25+o(x) on unbounded dyadic scales is sufficient, using the certified margin C2*(1-A2)>33/200. That discrepancy estimate remains OPEN. Vatwani's recovered author preprint clarifies the one-odd-exponent cases and unfilled near-level-one hypotheses. No new signed bound is obtained; literature novelty is unestablished. | [moving-cutoff-parity.md](moving-cutoff-parity.md) |\n| C | `Q-next-correlation-source-map` Is there a primary-source statement with stronger scale or coefficient quantifiers than the ones the prime-band transfer actually consumed? | ANSWERED | Scoped negative for the inspected statements: function-class and rate mismatches remain. The coordinating review withdraws the all-scales/dyadic-omitting requirement as a necessary gate, since bounded-interval stability yields a weak dyadic scale average from the existing input. No sufficient rate or full-coefficient theorem is matched. Proper-prime-power terms are already negligible on the fixed corner; other prime-r cofactor branches remain open. | [next-correlation-source-map.md](next-correlation-source-map.md) |\n| C | `Q-next-transfer-review` Does prime-band-transfer.md (1)-(6) follow at exactly its stated scope, and does any step fail against an adversarial reading of its hypotheses? | ANSWERED | The derivation (1)-(6) survives at its stated continuous-average scope. The coordinating round review corrects check l: interval stability additionally gives a dyadic scale average, with a further exponent loss. There is no every-dyadic estimate, o(N), full-corner control or twin margin. Proper-prime-power branches have a separate negligible fixed-corner bound. | [next-transfer-review.md](next-transfer-review.md) |\n| C | `Q-paired-factor-budget` Does averaging the proposed two-small-prime loss and three-small-prime gain using existing distribution theorems make their net contribution sufficient for the current twin consumer? | ANSWERED | DERIVED from the matched classical inputs, with finite rational enclosures VERIFIED: the explicit positive family has liminf P3/(C2*x)>1, but liminf (N2-P3)/(C2*x)>11/10. Keeping only these families and the main term therefore fails at the current cutoffs. Pan--Ding supplies aggregate level x^(9/20), stronger than the earlier per-cofactor application; the complete signed margin remains OPEN. The aggregation machinery is prior art and literature novelty of this application is unestablished. | [paired-factor-budget.md](paired-factor-budget.md) |\n| C | `Q-polylog-fold-transfer` Can the fixed-fold calculation be made uniform at growing depth, and transferred to the actual bilinear remainder with an adequate error? | ANSWERED | The local bilinear model is O_H(x/log^H x) uniformly for polylogarithmic squarefree moduli. Its absolute reconstruction error is at least (1/2+o(1))*x. The needed signed reconstruction bound remains OPEN, so this model does not establish the twin-prime lower bound. | [polylog-fold-transfer.md](polylog-fold-transfer.md) |\n| C | `Q-prime-band-completion` Can the remaining squarefree correlation be localized to simpler prime and harmonic bands, and does completion then permit a direct general bilinear Kloosterman estimate? | ANSWERED | The pilot reduces, with power-saving endpoint errors, to p in (x^0.235,x^0.24], q in (x^0.045,x^0.05] and positive h in (x^0.029,T], with T=O(x^0.0342). Both primes exceed 2h eventually. Completion is exact but produces coupled h-dependent coefficients on every dual residue modulo q. Its full-frequency Kloosterman matrix has norm exactly q, so a uniform power saving by an arbitrary-coefficient operator bound is impossible at that scope. This completion alone leaves the 61/20000 budget deficit; prime-dispersion.md controls the pilot using a different second moment. No twin lower bound is established. | [prime-band-completion.md](prime-band-completion.md) |\n| C | `Q-prime-band-transfer` Can the nonmultiplicative prime-band weight be represented by bounded multiplicative functions, and what does an available quantitative correlation theorem actually give after all scale and normalization costs? | PARTIAL | A smooth Fourier superposition represents mu(n)L_B(n)/log X exactly on n<=2X+2. The quantitative non-pretentious input and a mesh give a continuous scale-average saving for the exact s=s'=1 windows. The round review additionally derives a dyadic scale-average saving by bounded-interval stability, with exponent c_*/2. Neither estimate gives an every-dyadic bound, o(N), full-corner control or a twin margin. The extension in cofactor-progression-transfer.md now estimates a growing polylogarithmic cofactor family; the full remainder remains open. Proper-prime-power branches have the separate bound in corner-correlation §1.1. | [prime-band-transfer.md](prime-band-transfer.md) |\n| C | `Q-prime-detection-inputs` Can we give a locally correct positive comparison sequence, justify the ordinary arithmetic inputs, and prove exactly what one additional estimate would buy? | ANSWERED | Yes: the comparison sequence and Type I estimates are derived, and a classical Vaughan decomposition gives S(x)=C2*x+B(x)+O_H(x/log^H x) for every fixed H. A fixed one-sided saving over B>=-C2*x+o(x) implies twins on unbounded scales. This is a fully specified classical reduction, not a new cancellation estimate or proof mechanism. | [prime-detection-spec.md](prime-detection-spec.md) |\n| C | `Q-prime-dispersion` Can a second moment retaining two right primes control the remaining localized squarefree pilot after its actual conductor and coefficient costs are included? | ANSWERED | Yes at this rectangle: aggregation in m=dp before Cauchy gives a logarithmically bounded coefficient on a long interval. The only zero-phase pairs are identical (q,h). Distinct primes give modulus e*q1*q2 with gcd loss at most O(A*Z); completion and Weil bound all off-diagonal terms. The diagonal has exponent 1989/2000 and the largest off-diagonal exponent is 1173/1250 before an arbitrarily small loss. The remaining high-frequency polynomial is O(x^(199/200)); consuming the previous reductions gives R_IJ=O_H(x/log^H x) at D=x^0.277, E=x^0.467. This is a full rectangle at product scale x^0.744, not a uniform product cutoff or a twin lower bound. | [prime-dispersion.md](prime-dispersion.md) |\n| C | `Q-prime-power-dispersion` Does the repeated-prime structure control the BB contribution beyond its two-norm bound, and can a prime-power moment include every harmonic uniformly? | ANSWERED | The decomposition leaves an exceptional coefficient with norm at most sqrt(D) times logarithms; its odd prime-power core has a derived all-harmonic dispersion bound. The same bound for first powers removes the former left-prime cutoff and unpaired diagonal constraint. Two strict inequalities control a region including d~x^0.247, e~x^0.612 at product scale x^0.859. Their product-exponent supremum is 5757/6700, not a uniform product cutoff or a twin lower bound. Exact finite checks run separately from this classical-input derivation. | [prime-power-dispersion.md](prime-power-dispersion.md) |\n| C | `Q-reachability-coverage` Which part of the original divisor domain can no improvement to the grouped moment's nonzero kernel reach, in either orientation, and what would a uniform saving on that kernel actually buy? | ANSWERED | For a prescribed fixed exponent margin eta0, the zero budgets leave S_0={p<2eta0,q<2eta0} below that required margin regardless of the nonzero-kernel saving. Uniform gamma=2 covers its complement at that prescribed margin. Smaller positive margins can reach fixed interior points of S_0; only p=q=0 remains at zero budget one for every margin. The proposed eta0~loglog x/log x substitution is not justified by estimates carrying x^epsilon losses; its box count is geometry only. The sufficient twin margin remains OPEN. Independent review F3 and F5 correct the quantifiers and the prime-power tail proof; round-review-0906 corrects the diagonal scope, fixed-margin bulk claim and six-cut separation. | [reachability-coverage.md](reachability-coverage.md) |\n| C | `Q-research-execution` How should agents use the integrated returns, avoid repeated work and select the next bounded research obligation? | ANSWERED | The second dispatch from 1285d47 closed at 46fa948 and was independently reviewed on 2026-09-09. A/A2/B/C/D/E and V/W returns are present and integrated at their owning scopes. A2's corrected Type I estimate is accepted; D's fourth residual cut is retained and its remaining moment saving corrected to 7/200. C's low nonzero modes and E's aggregate scope are corrected. No new assignment is started by this integration; next obligations are candidates, not running work. Twin-prime infinitude and every sufficient signed margin remain OPEN. | [RESEARCH-EXECUTION.md](RESEARCH-EXECUTION.md) |\n| C | `Q-research-handoff` What must an incoming research agent understand, verify and return before its work can advance the current twin-prime argument? | ANSWERED | Current handoff after the 2026-09-09 independent review of the dispatch ending at 46fa948. The full residual has four simultaneous complement conditions after D's regional estimate. A's centered truncation and A2's repaired low Type I estimate give the alternative S=C2*x+B+O_A(x/log^A x). B and all sufficient signed margins remain OPEN. The C3 absolute bound, family identities and scoped C/E failures retain their owning hypotheses. Execution is owned by RESEARCH-EXECUTION.md. | [RESEARCH-HANDOFF.md](RESEARCH-HANDOFF.md) |\n| C | `Q-research-round-validation` Which submitted results from lanes A, A2, B, C, D, E, V and W survive direct review, and what can be integrated or used to choose the next attempt? | ANSWERED | Reviewed incoming commits 07ab47f through 46fa948 on 2026-09-09. A's truncation and A2's repaired low Type I estimate are accepted with bookkeeping corrections. B's ordinary-BV family identity survives; its uniform Dickman constant and complement condition are repaired. C's norm obstruction survives on all smooth inputs, but low nonzero Fourier modes are also 1-bounded. D's block estimate and fourth residual cut survive; its next moment saving is 7/200, not 17/200. E's rough-product distribution and aggregate constant 4 are accepted at the stated Wu uniformity reading; rational certificates close the two displayed tests for every fixed u>4. W remains OCR-only at the original 1974 theorem page. All sufficient twin-prime margins remain OPEN. | [research-round-validation.md](research-round-validation.md) |\n| C | `Q-residual-coverage` What portion of the full residual is controlled with uniform margins, and can the exceptional first branch improve that coverage rather than just a regional corner? | ANSWERED | The first-branch and prime-free moments remove the exceptional norm condition. Full rectangles are controlled when delta+nu<19/25 or 5delta+2nu<123/50. Uniform twists justify exact coupled cuts: every fixed de<=x^theta with theta<19/25, and d^5*e^2<=x^lambda with lambda<123/50, as well as their union, have arbitrary fixed logarithmic savings. A concrete remaining endpoint sum has de>floor(x^(3/4)) and d^5*e^2>floor(x^(49/20)). Its twin margin is OPEN. The right zero-frequency budget, after an extra Cauchy inequality, limits the uniform product range; the left cross moment limits the other edge. Finite checks validate identities and masks, not asymptotic rates. | [residual-coverage.md](residual-coverage.md) |\n| C | `Q-review-request-0906` Which claims of the arithmetic campaign should an independent reviewer check, in what order, against which records, and what must the review return? | ANSWERED | Completed review specification at the bounded scope recorded in handoff-review-0906.md and reports 20 and 21, with the subsequent round reviewed in round-review-0906.md. The dependency list records questions at dispatch, not current unfinished assignments. Imported proofs retain their review limits and the twin margin remains OPEN. Current assignments are in RESEARCH-EXECUTION.md. | [REVIEW-REQUEST.md](REVIEW-REQUEST.md) |\n| C | `Q-round-review-0906` Which conclusions of the completed agent round survive review, and which corrections change the next research specification? | ANSWERED | The fixed-band lift and continuous moving-window estimate survive the checked source interface. Interval stability additionally gives a dyadic scale-average logarithmic saving, so the blanket scale exclusion was too strong. The rate and missing prime-cofactor terms with s>1 or s'>1 still prevent a twin margin. Proper-prime-power branches already have a negligible bound on the fixed corner. The driver computes integer parity, not the CRT gcd partition; its total coefficient identities survive. Diagonal lower-bound, reachability and six-cut bookkeeping overclaims are corrected. No region, exact residual cut or twin lower bound changes. | [round-review-0906.md](round-review-0906.md) |\n| C | `Q-sharp-corner-transition` What is the energy of the complete sharp corner coefficient, and can the smoothed estimate be transferred through a negligible L2 transition? | ANSWERED | DERIVED using named analytic inputs: for each fixed 00 both sharp squared norms and sharp-minus-smoothed squared norms are Theta_eta(x log^2 x) on the actual dyadic intervals. Thus an O_eta(x log x) sharp squared norm or negligible L2 transition fails in that range. Signed transition correlations and the global complement remain OPEN. No region, exact residual cut or twin margin changes. | [sharp-corner-transition.md](sharp-corner-transition.md) |\n| C | `Q-shifted-prime-decomposition` Does decomposing Lambda(dk-2) give a provable saving on any part of the actual shifted-prime sum, and what exact arithmetic remains outside the imported hypotheses? | ANSWERED | Both second Type I terms are O_H(x/log^H x) by classical Mobius BV. The residual is an explicit weighted sum over dk-ev=2, equivalently an average of two-linear-form Mobius correlations with growing coefficients and possibly one-point intervals. Its required one-sided improvement is OPEN. No twin lower bound or novelty is claimed. | [shifted-prime-decomposition.md](shifted-prime-decomposition.md) |\n| C | `Q-shifted-prime-mobius-sums` At reachable x, do the shifted-prime Mobius and Liouville sums, the cleanest instance of the parity input the twin reduction lacks, show any size or sign structure that a random-sign model does not, and how loose is the trivial bound on M(x) used for the centered tolerance? | ANSWERED | MEASURED to x=2^38 (10.9e9 primes). All four sums sit at random-sign size: \\|U\\|/control rms at most 2.8 at any j, slopes 0.28 to 0.65 against the control's 0.43 to 0.47 with per-draw spread 0.19 to 0.30; the pre-registered structure falsifier did not fire. M(x)/x is at most 8.8e-5 in magnitude for j>=29 against the trivial bound 0.374, so the -4x/25 tolerance in moving-cutoff-parity is loose by four orders at these scales; a proof still needs a bound on M, which is the same parity object. The squarefree density of p+2 among primes is 0.747911 at 2^38 against 2*Artin = 0.7479116. pi(2^j) matches OEIS A007053 at every j. Nothing here bears on the asymptotic conjecture. | [shifted-prime-mobius-sums.md](shifted-prime-mobius-sums.md) |\n| C | `Q-signed-divisor-grouping` Can grouping different fibers while retaining their Mobius signs control an explicit part of R, and what signed error remains after that grouping? | ANSWERED | For every fixed H>0 the actual remainder restricted to de<=x^(7/10) is O_H(x/log^H x), although each sign's mass there is at least c*x*log(x) eventually. This follows from a written classical-input derivation, with finite algebra and archived-factor checks. The remaining de>x^(7/10) contribution reduces to an explicit weighted CRT endpoint discrepancy whose required one-sided improvement is OPEN. | [signed-divisor-grouping.md](signed-divisor-grouping.md) |\n| C | `Q-signed-moment` Is there an argument for the block that does not pay sqrt(M) at the first Cauchy inequality or that extracts cancellation from the R=0 class, and what are its budgets at the corner a=b=1 and at the target box (delta,nu)=(8/25,9/20)? | PARTIAL | The true diagonal has an upper bound x^(1+o(1)) on the transition band, not a uniform nonzero lower bound. Lemma A controls only u1=u2,h1=h2; unequal proportional R=0 pairs remain outside it. Lemma B and the now-priced joint Cauchy arrangement add no region. The former general Holder floor is valid only under additional coefficient assumptions, not for arbitrary sparse coefficients. Conditional off-diagonal budgets and the random-matrix ceiling remain conditional and heuristic. No necessity of using both signs, full-corner estimate or sufficient twin margin is established. | [signed-moment.md](signed-moment.md) |\n| C | `Q-singleton-fiber-audit` Can the one-point fibers of the coupled remainder be discarded or controlled in absolute value at scale x, and what do the actual weights and archived intervals show? | ANSWERED | The grouped coefficient is zero whenever either integer is prime. Each sign's ungrouped singleton mass is at least c*x*log(x) eventually for an absolute c>0, by a written derivation from named classical inputs. Finite full-interval and archived-prefix checks validate the partition. The signed singleton and longer-fiber contributions remain OPEN; no twin lower bound or novelty is claimed. | [singleton-fiber-audit.md](singleton-fiber-audit.md) |\n| C | `Q-small-divisor-kernel` Does reciprocity separate the j<=x^(1/20) kernel of grouped-divisor-moment (21) at delta=8/25, nu=9/20, and do the Duke-Friedlander-Iwaniec / Bettin-Chandee and Kuznetsov interfaces then supply the required saving? | ANSWERED | The separation is exact and cheap: reciprocity splits index by index, the correction phase has sup norm and total variation O(A/(MN))=O(1/x) at this box, and Mellin separation of the endpoint weight costs x^epsilon with L1 mass min(1,Ax/(MN)), so the whole block reduces to a pure trilinear Kloosterman fraction with unimodular twists. The interfaces then fail by explicit amounts: Bettin-Chandee Theorem 1 gives block exponent 129/125, above the required 1 by 4/125 and above the classical 103/100; DFI (1.1) gives 1267/1200; every box Bettin-Chandee controls is already controlled, verified on 19005 rational boxes; the audit's spectral diagnostic factor is trivial for all common divisors up to x^(19/50). Closing this box inside the same interface needs the (M+N) exponent below 27/140 or the (AMN) exponent below 9/28. Target (21) and the global twin margin remain OPEN; nothing is added to the controlled region. | [small-divisor-kernel.md](small-divisor-kernel.md) |\n| C | `Q-smooth-sieve-literature` Which nearby published results should be imported instead of rediscovered, and do their methods give a more focused next question for the complete global remainder? | PARTIAL | Existing sources supply the finite-difference mechanism, one-point concentration and quadratic profile optimization; shifted divisor-sum estimates admit fixed shift two under their own support and smoothness conditions. A derived application of corrected Henriot bounds gives small-prime mass O_sigma(delta^sigma x)+O_epsilon(x^(39/40+epsilon)), uniformly in delta for fixed 0floor(x^(77/100)) as a fourth simultaneous condition of W_dagger through the Perron machinery, changing only the summation domain of E_dagger. The target box (407/400), the corner, the uniform product threshold 19/25, the sign of E_dagger and the global twin margin are unchanged; the margin remains OPEN. | [structured-dispersion-estimate.md](structured-dispersion-estimate.md) |\n| C | `Q-supported-coefficient-dickman` Can the whole supported coefficient be priced using existing literature before enumerating its factor cells, and does the resulting separate-sign sieve estimate supply a positive twin margin? | ANSWERED | DERIVED from classical inputs: at delta=1/100 the harmonic limit is H_delta=D(25/12)-1+E_delta with 0<=E_delta<=24/22!, hence -13/2, where I is the explicit ordered two-factor integral below. A three-prime left input and rough composite right input suffice; prime partners are subtracted using an upper sieve. The witness survives every fixed delta<1/50 and every averaging profile with the same plateaus. Consequently the negative-only consumer is false, and liminf Phat/(C2*x)>1/2. The signed twin margin remains OPEN; literature novelty is unestablished. | [switching-negative-mass.md](switching-negative-mass.md) |\n| C | `Q-transition-energy-review` Does the sharp-energy chain in sharp-corner-transition.md survive an adversarial check, and do the upper norm, the small-fixed-eta lower norm and the failed negligible transition each stand at their stated scope? | ANSWERED | All ten assigned rows survive at their stated scope; no counterexample or decisive contradiction was found. The upper norm, the small-fixed-eta lower norm and the refuted negligible L2 transition each stand separately. Two repairs are citation-level only: the integer Barban-Vehov input is Graham 1978 quoted by Chen An, and the note should record why the k=Q constant is absolute. The de la Breteche-Dress-Tenenbaum PDF was refetched and its SHA-256 matches the pinned hash; its (1.5) and Theorem 1.1 supply exactly the imported (3) and (4). eta_0 remains positive and unextracted, and the signed sums, E_out and the twin margin remain OPEN. | [transition-energy-review.md](transition-energy-review.md) |\n| C | `Q-transition-energy-review-spec` What independent checks are required before the sharp-energy and non-negligible-transition conclusions can guide the next signed research attempt? | ANSWERED | Bounded review assignment with explicit source, algebra, interval, error and quantifier checks; per-claim verdicts and dependency consequences are required. The assignment is specified, not executed. | [transition-energy-review-spec.md](transition-energy-review-spec.md) |\n| C | `Q-transition-joint-budget` Can the mixed and smoothed terms be handled jointly with the transition pair, and what complete inequality would make such an estimate useful for the twin consumer? | PARTIAL | The exact split, cutoff-average identities and separate upper budgets survive. Averaging cutoff norms after triangle and Cauchy is saturated for sufficiently small fixed eta, but signed cutoff arguments and joint cancellation are not closed. R_00 need not be estimated separately under every grouping. The short-window constraint is corrected with a bounded majorant and a safely reduced lower constant. No consumer input or signed improvement is supplied; E_out and the sufficient twin margin remain OPEN. | [transition-joint-budget.md](transition-joint-budget.md) |\n| C | `Q-transition-joint-budget-spec` Can the mixed and smoothed terms be handled jointly with the transition pair, and what complete inequality would make such an estimate useful for the twin consumer? | ANSWERED | Bounded assignment to audit the exact four-term decomposition, price a concrete joint treatment, and return a sufficient one-sided or common-scale inequality with all open terms explicit. The decomposition and existing norm budgets are inputs, not new findings. | [transition-joint-budget-spec.md](transition-joint-budget-spec.md) |\n| C | `Q-transition-research-execution` How should the core research handler execute and integrate the three bounded transition assignments while retaining the central signed estimate? | ANSWERED | Prepared execution contract with a fixed mathematical baseline, three independently assignable specs, handler-owned signed research, dependency checks, shared-file ownership, reporting requirements and stopping rules. Execution is complete; the corrected results are in transition-round-review.md and transition-round-audit.md. The specification itself supplies no arithmetic estimate. | [TRANSITION-RESEARCH-EXECUTION.md](TRANSITION-RESEARCH-EXECUTION.md) |\n| C | `Q-transition-round-audit` Which conclusions of the completed transition round survive direct review, which require correction, and what research decisions remain justified? | ANSWERED | The sharp upper norms and sufficiently-small-fixed-eta lower norms survive. The integration contained substantive errors: an omitted lower cutoff, an incorrect Mellin formula for the transition, a conflation of singleton divisor fibers with long cofactor fibers, an unpriced source representation and overbroad closures of joint or cutoff arguments. Owning notes and current records are corrected. No signed improvement or sufficient twin margin is established; the corrected cofactor and joint-correlation interfaces remain open. | [transition-round-audit.md](transition-round-audit.md) |\n| C | `Q-transition-round-review` What survives the executed transition round after direct review, and what combined estimate is still missing? | ANSWERED | The energy chain survives at its original scope; the integration's missing cutoff, Mellin representation, fiber identification, normalization and general closure claims are corrected. The exact cofactor and joint-cutoff formulations remain useful. No signed estimate, controlled region, exact cut or sufficient twin margin changed. Only the stated norm-only procedure and the arbitrary-factor fixed-shift relaxation are refuted at their specified scopes. | [transition-round-review.md](transition-round-review.md) |\n| C | `Q-transition-signed-estimate` Can the signed transition kernel (E1) be estimated beyond its absolute norm budget, and if not, what exact input does the consumer need? | PARTIAL | No signed improvement. Derived: the proper-prime-power part of R_11 is O(x^(39/40+epsilon)); the prime part is an exact beta-weighted sum over pairs of linear forms of two-point Mobius correlations with coefficients up to x^(6/25+2eta) and x^(1/20+2eta); the Cauchy budget is at most 2B*eta*x*log^2(x)(1+o(1)) and the fiber triangle budget 4*eta^4*x*log^4(x)(1+o(1)), so every existing inequality loses one of the two signs. The cutoff-difference identity converts T at pn into a sharp-window coefficient times log(p)/L plus two boundary windows; the entropy-decrement route then needs bounded coefficients, a short-interval one-point input for T and a log-weighted consumer, none available. The required one-sided input is stated exactly in section 5. The twin margin remains OPEN. | [transition-signed-estimate.md](transition-signed-estimate.md) |\n| C | `Q-transition-source-match` Do the inspected divisor-correlation, shifted-convolution and averaged-Chowla statements directly supply a usable estimate for the exact transition kernel? | ANSWERED | No direct supplier was established among the inspected statements. The cofactor exchange and Theta identity are exact, but their scalar coefficient mass is not a full bounded-class representation for T. The transition Mellin kernel requires a sharp subtraction with a 1/s tail. Fixed-divisor singleton fibers differ from potentially long cofactor fibers. The fixed-shift arbitrary-factor relaxation is refuted; broader signed and cutoff methods remain open. | [transition-source-match.md](transition-source-match.md) |\n| C | `Q-transition-source-match-spec` What bounded source investigation could supply a usable signed estimate for the transition kernel, and how must its hypotheses and losses be priced? | ANSWERED | Specification for up to three detailed source-interface assessments, including coefficient representation, scale quantifiers, normalization, separation errors and exact downstream payoff. No new theorem match or literature-absence result is claimed. | [transition-source-match-spec.md](transition-source-match-spec.md) |\n| C | `Q-twin-reduction` What has the arithmetic campaign of 2026-09-05 and 2026-09-06 actually established about the twin sum, at what calibration, what exactly remains, and what would suffice? | ANSWERED | Consolidates the checked regional reduction S=C2*x+E_dagger+O_H(x/log^H x), its sufficient unbounded-scale margins, coefficient definitions and source dependencies. Later complete smooth and centered formulations are owned by RESEARCH-HANDOFF.md and their linked derivations. No new arithmetic estimate is supplied. Imported theorem proofs retain their stated review limits; every sufficient twin margin remains OPEN. | [TWIN-REDUCTION.md](TWIN-REDUCTION.md) |\n| 9 | `Q-applied-0828-varE` Were the Var/E red team's corrections applied to the four HELD notes? | ANSWERED | Every WEAKENED and REFUTED verdict in redteam-0828-varE.md is applied to its target note in the present tense; four ledger verdicts are rewritten, the theta=1 row is raised to PROVEN, and seven live-doc sentences plus one producer reading stay open and are drafted here. | [applied-0828-varE.md](history/staging/applied-0828-varE.md) |\n| 9 | `Q-lit-scourfield-2008` Does Scourfield, Smooth divisors of polynomials (LMS Lect. Note Ser. 352, CUP 2008, 286-311), supply the divisor-distribution estimate that varE-theta2-proof.md section 6 leaves open? | ANSWERED | DOES NOT DELIVER, at SECOND-HAND rung, and the chapter is still UNREAD at the page: ten channels were attempted, nine answered, and every one that answered is closed, degraded or paywalled, but the author's own restatement of the 2008 theorem, read at the page image of Scourfield, Funct. Approx. 55.1 (2016) 84 eq. (1.1), counts divisors m <= x of f(n) for n <= x with weight 1, for f a product of pairwise coprime irreducible factors each of degree at least 2, and the open step needs divisors n > 2L above the window, carrying a branch-dependent weight of total mass 4^omega, for the product of three LINEAR factors C(C-2)(C+2). Two gaps each independently fatal, plus a third that is AMBIGUOUS between two author-sourced statements and does not carry the verdict. One correction to lit-smooth-divisors.md is earned: the precision register of the owning convention is not uniformly (log y)^{o(1)} or order-of-magnitude, since (1.1) is an asymptotic with relative error O(1/log x), which is the right register and one epsilon short of the o(1/ln y) the open step needs. Step 1 of Var/E does NOT become conditional on a citation, and no absence may be written, since the chapter's interior lemmas were not read. | [lit-scourfield-2008.md](history/staging/lit-scourfield-2008.md) |\n| 9 | `Q-paper-iii-restructure` Is Paper III restructured around the proven model theorem? | ANSWERED | Yes. paper/variance-note.md now carries four new sections in place of its fit narrative: sec.8 the proven spectral form (Var/E = delta X, delta ln^2 W -> 16 C_2 e^{-2gamma}/3 = 1.109905, the Montgomery-Soundararajan main term exactly L prime by prime, so all of Var/E is discrepancy), sec.9 the decoupled model with Theorems 7 and 8 and Lemmas 9-11 written out with proofs and the closed form lambda_2(2) = 1 - e^{-2gamma}(9/2 - 4 ln 2) = 0.45546 plus the theta = 1 identity with Gorodetsky, sec.10 Conjecture 1 (delta(X - X_dec) -> 0) with eight exact ratios, two of three lag groups closed and the third named as a divisor-distribution statement not yet searched, and sec.11 the 0.611 refutation as a short calibration paragraph; sections 1-5 are untouched, the abstract and section 6's open question are rewritten to match, and the \"u = 2 exactly\" defect in section 7 is corrected. | [paper-iii-restructure.md](history/staging/paper-iii-restructure.md) |\n| 9 | `Q-recon-0830-smooth-aps` Does any theorem on y-friable integers in arithmetic progressions (Granville, Fouvry-Tenenbaum, Soundararajan, Harper, Drappeau, Drappeau-Granville-Shao) supply the uniform o(1) equidistribution that attack-0830-varE-identification.md section 4 leaves open, for moduli up to y^(4/5) with the lam1 weight and the squarefree restriction? | PARTIAL | NONE APPLIES AS STATED, and the step stays open; lim Var/E = 0.45546 stays HEURISTIC. Read at the page (eleven sources, arXiv text layers and the Acta Math. page image): every pointwise asymptotic is O(log q / log y)-precise (Granville 1993 Thm 1) or hypothesises log x / log q -> infinity, and every beyond-square-root level (Fouvry-Tenenbaum 1996, Drappeau 2015, Drappeau-Granville-Shao 2017, Pascadi 2023/2025) hypothesises y <= x^delta, i.e. u -> infinity, against our u in (1.2, 2]. NEAREST: Harper arXiv:1208.5992 (2012, a preprint with no journal version found) Theorem 2, the Barban-Davenport-Halberstam form, whose range (log^K x <= y <= x, Q <= Psi/log^A x, bounded u in the ineffective form) covers d up to L^(1/2)/log and whose weight (1 on all friable n, no squarefree restriction, no max over x' <= x) is unmet. The record's quantifier is also refined: the cell needs the lam0-weighted AVERAGE over d, not uniformity. PROVEN given the weighted form (H_w) of that theorem, derivation mine and unreviewed, with its two identities checked numerically in the producer: every two-branch cell of Xmix, both mirror cases, all bands, balanced blocks included, is o(ln^2 y), so the Bettin-Chandee kernel separation is not on this route; the reduction of (H_w) to Harper's theorem is OUTLINED and not written, and the three-branch type (measured -0.0002 at x = 19, trivial bound O(ln^3 y)) is untouched. Not TPC-strength. | [recon-0830-smooth-aps.md](history/staging/recon-0830-smooth-aps.md) |\n| 9 | `Q-redteam-0828-varE` Does the 2026-08-28 chain from the exact comb variance to lim Var/E = Pr[GD(2) > 2] = 0.45546 survive an adversarial re-derivation, and does the refutation of the 0.611 reading hold? | ANSWERED | The constant survives at HEURISTIC and the 0.611 refutation is CONFIRMED and strengthened (the frozen out-of-sample half of the protocol also fails on the control); the correction is that TWO steps are open, not one, and that the theta=1 branch is an exact identity with a read theorem rather than a second-hand numerical match. | [redteam-0828-varE.md](history/staging/redteam-0828-varE.md) |\n| 9 | `Q-redteam-0830-imports` Do the six lighter notes of 2026-08-30 (varE-identification, smooth-aps, rec-killrun, skeleton-door, at43-bigint, coherence) survive an adversarial pass on independent code and at the source pages, and which of their load-bearing claims are wrong? | ANSWERED | No note is refuted whole and no exponent moves. Every decisive measurement reproduces on code sharing nothing with the producers: the varE mixed remainder and its 0.844 / 0.826 share below 2L and 94.4 / 98.7 percent unbalanced share at x = 13, 17; the Buchstab deviations and all three crossing levels log10 z* = 33.53, 68.80, 33.62; the skeleton door's 9.2 / -0.9 percent, G30_agg 0.1113 / 0.1011 and the 0.0102 cap at @13, @17, by a path that also verifies cap-36 Theorem A; the @43 parity exactness on all 26,157,448 scour classes with 0 above 2^54 and a constructed @47 counterexample; and the five recon-0828 verdict counts summing to 75, which REFUTES rec-killrun's brief-error clause and confirms the orchestrator's rider. Seven claims WEAKEN. The load-bearing one is smooth-aps's \"the weight is unmet in print\": Harper's J. London Math. Soc. 112 (2025) e70293 proves a Barban-Davenport-Halberstam asymptotic for an ARBITRARY sequence on exactly that dyadic all-classes object, and his own footnote says the missing max over x' can be incorporated, so the NEAREST citation changes even though the step still does not close (the 2025 range is sqrt(2x) < Q <= x and its Theorem 2 route bars sieved sets by name). Also: Henriot Corollary 2 applies, but to the ERRATUM's statement and under two hypotheses the note leaves unstated (eps < alpha/600, not \"every eps\"; F at prime powers); varE section 4's display (*) carries an unfilled placeholder and is vacuous by a factor L/d; the skeleton door's \"would tighten\" is wrong in sign at @17, where the closable block is negative and removing it RAISES G30_agg to 0.1020; Halasz-Montgomery is not applicable for the reason given (it is an abstract inner-product inequality); FGKT's heuristic quote is on p. 4, not p. 3; and the \"~13x had no source\" is a corpus miss, the retired G2-walk figure at CHANGELOG.md:5716. Not TPC-strength; nothing here opens a route. | [redteam-0830-imports.md](history/staging/redteam-0830-imports.md) |\n| 9 | `Q-redteam-0830-slack` Do attack-0830-head-remainder.md, attack-0830-tail-derivation.md and verify-0830-record-defects.md survive an adversarial re-derivation on independent code, and do the five riders the verification put on live notes stand as written? | ANSWERED | Arithmetically they survive: every figure of the three notes that this pass could recompute reproduced to the printed digit on code written from the definitions, 0 assertion failures over 58 assertion call sites (most inside per-level loops), including every figure of the verification's three claims and the half-decade Delta_HL it quoted rather than recomputed. Four sentences do not survive. (1) verify-0830-record-defects.md:316, a replacement queued for head-residual-hl3.md sec.0, says the pooled value sits \"0.038 to 0.052 below the sub-window mean at all three decades\"; measured, the two lower decades run to 0.076 and 0.066, and the same note's own falsifier row says 0.038 to 0.076, so the wrong half is the one queued to land. (2) verify-0830-record-defects.md:379's un-measured \"+0.005\" residual inside the eighths is +0.0004 measured at sixteenths, so the de-pooled top-decade value is converged and the caution can be dropped. (3) the ruling that the class null is \"the matched figure\" for the tail is matched on residue class only: p'^2 is coprime to every q <= p', and the ensemble's rough-class offset is E_rc - R = 5.6714 -> 7.5357 over x = 7..29 against the class null's 5.6000 -> 6.0336, giving t/(R_shell + 7.5357) = 1.0123 against t/classNull = 1.0157. (4) attack-0830-head-remainder.md sec.3's top row is not gap-scale matched: E[g] 244.0 against 235.9, y_match capped at 19997, and on the note's own six ensemble points Delta_ens ~ E[g]^-0.312, so meas/ens reads 0.9640 not 0.9539 and the headline \"0.954 to 0.995\" reads 0.964 to 0.995. Three levels the notes recorded as out of reach are run here: the X2 group at x = 23, the tile identities and conditionals at x = 29, and the ensemble ladder at y = 29 (pipe/ens 0.9535, 40 s, against the head note's \"about fifteen minutes\"). No route opens or closes and nothing here touches Z2. | [redteam-0830-slack.md](history/staging/redteam-0830-slack.md) |\n| 9 | `Q-var41` What does the stable law predict for Var(41), and what can the tenth Var/E point pin? | OPEN | Pre-registration only, sealed and committed alone before any Var(41) engine exists: it freezes the prediction, a band taken from the law's own residuals at z <= 37, the derived z(41) prediction, and the honest statement that one more point cannot separate a limit from a drift. | [var41-prereg.md](history/staging/var41-prereg.md) |\n| 9 | `Q-varE-identification-0830` Can either open step behind lim Var/E = 0.45546 (the identification delta*(X - X_dec) -> 0, or the theta = 2 mean-coefficient replacement) be proven, and if not, which single inequality does not close? | PARTIAL | Neither closes, and the two named steps are one statement (Conjecture 1 of variance-note sec.10; the model half is PROVEN in varE-limit-theorem.md). The remainder is re-split exactly at five levels: the corpus's X2 column doubles the positive half of one shift pattern (X2c = 2.66 against the group sum 0.09 at x = 19, PROVEN by rebuild), so its Xmix is overstated 2.3x; the true mixed remainder is MEASURED at -0.49 ln y and 82 % of it sits BELOW 2L, in the two-branch type, in the unbalanced range min(d,e) <= L^(2/5) (99.7 % of that cell at x = 19), outside the range of every bilinear Kloosterman-fraction bound; the divisor-distribution estimate the record names (n > 2L only) is therefore not the whole open step. Henriot's Corollary 2 (read at the page, erratum read) applies as stated and cuts the moduli to n <= L ln^(2+o(1)) y, PROVEN given that theorem; the one inequality left is the uniform o(1) equidistribution of y-friable squarefree integers in progressions to moduli up to y^(4/5), NOT SEARCHED at the page. The limit 0.45546 stays HEURISTIC. Not TPC-strength. | [attack-0830-varE-identification.md](history/staging/attack-0830-varE-identification.md) |\n| 9 | `Q-varE-limit` Does lim Var/E on the diagonal window exist, and what is it? | PARTIAL | The theta=2 mean-coefficient step is neither proven nor refuted: the replacement error is exactly a sum over shifts of W(h) - V(h), it splits into a c=0 group (needs no decoupling) and CRT-mixed lags (open); measured ratios true/model 1.0098, 1.0039, 1.0014, 1.0013 at x = 13..23 with delta(X - X_dec) ln y falling rather than settling, so the error is O(1/ln y) or better MEASURED on four levels that exclude growth and nothing finer, and the limit 0.45546 stays HEURISTIC with varE-spectral's second step, its own limit theorem, still open. | [lit-dickman-variance.md](history/staging/lit-dickman-variance.md), [lit-smooth-divisors.md](history/staging/lit-smooth-divisors.md), [varE-asymptotic.md](history/staging/varE-asymptotic.md), [varE-exact-ladder-01.md](history/staging/varE-exact-ladder-01.md), [varE-limit-theorem.md](history/staging/varE-limit-theorem.md), [varE-spectral.md](history/staging/varE-spectral.md), [varE-theta2-proof.md](history/staging/varE-theta2-proof.md), [varE-theta2-step.md](history/staging/varE-theta2-step.md) |\n| 9 | `Q-verify-record-defects-0830` Do the three defect claims of 2026-08-30 (attack-0830-varE-identification.md sec.8 on the doubled X2 column; attack-0830-tail-derivation.md sec.7 on the corpus's \"R + 1/2\"; attack-0830-head-remainder.md sec.1 on the decade-pooled Delta = 0.6214) reproduce on independent code from the definitions, and which of the corrections they list apply? | ANSWERED | Claim 1 CONFIRMED: the corpus delta*X2 column is the positive half of the p \\| h-2 pattern doubled, that pattern's mirror is the p \\| h+2 pattern (W-(-h) = W+(h) exactly, W-(h) != W-(-h) at 9 to 900,679 shifts), the group sum 2X2 reads -0.5603 .. 0.1846 at x = 7..19 against 5.7147 .. 5.3266, and Xmix is overstated 2.305x at x = 19 (6.9x at x = 7); X, X1 untouched. Claim 2 AMENDED: R + 5/2 and R + 3 are exact for the tail convention (all and odd origins, asserted at x = 7..23) and t/(R + 3) = 1.0228, but head-residual-factor.md:70 is the HEAD's forward convention where R + 1/2 and R + 1 are exact, so that correction does not apply; and p'^2 is 1 mod 6, where the tail constant is 5 (t/(R + 5) = 1.0181), with the class null (1.0157) the matched figure, so 1.0228 is one unmatched convention replacing another. Claim 3 CONFIRMED as an artefact, AMENDED on mechanism and on the half-decade reading: 0.6214 pooled against 0.6592, 0.6705, 0.6736 at 2, 4, 8 sub-windows; the between-slope (Simpson) piece the note derives is half of the pooling term (0.0187 of 0.0378 at halves, 0.0261 of 0.0522 at eighths), the other half is the within slope's own fall with height weighted by Var(g); and re-reading the halves from their eighths puts HL BELOW the measurement at 5 of 6 half-decades by 2.5 to 3.4 percent at the top decade, so \"within 1.6 s.e., sign alternating\" is itself a pooled statement. | [verify-0830-record-defects.md](history/staging/verify-0830-record-defects.md) |\n| Z2 | `Q-X-upper-0829n` Can the rough-pair census X(K) on the stretch window be bounded above, unconditionally and with an explicit error term, by a dimension-2 upper-bound sieve, and how loose is the bound against exact X(K)? | PARTIAL | Yes, as a theorem with nothing open (Selberg's Λ² inequality with a remainder bounded by the interval structure alone, VERIFIED at all 10908 anchor-depth pairs), but it is loose by 7.82 to 10.13 against exact X(K) at the band depths, decomposed as a sieve loss 1.78 to 2.53 times a structural loss 4.01 to 4.39 that no upper-bound sieve can remove, the window holds the level below z² (s_max 1.81 to 1.96) so the bound stops improving past p_K = 31 to 61, its main term is the sifting function's and not the census's, and the drift is a crossing quantity outside the legal half and is not evaluated. | [attack-0829n-X-upper.md](history/staging/attack-0829n-X-upper.md) |\n| Z2 | `Q-applied-0828-quadpoint` Were the quadpoint red team's corrections applied to the four HELD notes? | ANSWERED | Applied: the nine REFUTED and WEAKENED verdicts that target the four notes are written into them in present tense, with ledger blocks added; the tenth targets the sealed prereg, outside the fence, and is recorded in attack-quadpoint-02.md instead; three corrections owed by TODO.md Z2 and one by Z0 are listed here and are NOT applied. | [applied-0828-quadpoint.md](history/staging/applied-0828-quadpoint.md) |\n| Z2 | `Q-capture-identity` Does the transplanted cap family collapse onto a classical object, and what does the collapse say about the depth law and the wall? | ANSWERED | PROVEN, elementary: floor_K = T − X(K) on the half-open window with both members in [Q², Q′²), a convention that is load-bearing; the wall is renamed, not weakened; Σ capU counts each member against its own threshold, not against a uniform p_K, the two differing 11% to 37% at K*. | [quadpoint-identity-01.md](history/staging/quadpoint-identity-01.md) |\n| Z2 | `Q-derive-0904-r0-extension` Which pieces of the R0 decomposition (the tile form, the zone form, head, tail, Z2, and the composites R0/R1/R2/stretch) carry the parity obstruction's forbidden extension, and is any single piece both legal on the TPC-strength axis and exempt from it? | ANSWERED | No piece of R0 is both legal and exempt with content: the criterion reduces to a coordinate count (Proposition A) under which every statement forcing a twin pair into a range where roughness implies primality is covered, and the exempt residue is exactly the statements that assert no such forcing (R0's identity, lower bounds, boundary artefacts) plus the tile form, whose exemption is a property of the statement and provably does not close under deduction (Proposition B), so it survives only for routes with no sieve weight anywhere. | [derive-0904-r0-extension.md](history/staging/derive-0904-r0-extension.md) |\n| Z2 | `Q-parity-adversary` Does a parity adversary built on our own stretches survive at every level, and what precision class does a certificate need? | ANSWERED | It does NOT survive: the exact-data adversary dies at D* about Q^1.18 (MEASURED, exact arithmetic, 43 anchors Q <= 200), and the certificate that kills it is itself destroyed by 0.30 of one count per modulus, so Z2's precision class is now measured rather than described; it is PROVEN not to be a proof ingredient. | [attack-parity-adversary.md](history/staging/attack-parity-adversary.md) |\n| Z2 | `Q-parity-dstar-law-0829n` Does the exact-data parity adversary's killing level D*(Q) follow D* ~ Q^c with a stable c on the decades past Q = 200, and is c distinguishable from the independent-thinning null's? | CLOSED | No law in Q and CLOSED by the sealed kill rule (score 5 HIT, 1 MISS): MEASURED on 271 anchors 211 <= Q <= 13679, the Q-slope moves 1.446 to 1.204 to 0.575 across ranges while ln D* on ln width holds slope 1.097 +- 0.012 with a quarter of the scatter, D*/width about 0.5, and the independent-thinning null reproduces the width law within 0.025 in slope; the mechanism is that D* is where a modulus first reads a single position of the window, which is the wall survey's remainder statement in the stretch coordinate and names nothing new. | [attack-0829n-parity-dstar.md](history/staging/attack-0829n-parity-dstar.md) |\n| Z2 | `Q-quadpoint-blind-31607-0830` Does the zero-parameter depth law y*(Q), stated on the set where T(Q) >= 1 and on both the 1/(2e^gamma) = 0.280730 and the u*omega(u) = 2 (1/u* = 0.280438) forms with and without the finite-size Mertens factor, survive a sealed blind test on the bands past Q = 10007, and can that decade tell the two asymptotes apart? | ANSWERED | MEASURED, blind, and it decides nothing asymptotic: on 8363 new anchors 10009 <= Q <= 100003 the 76 sealed rows score 64 HIT and 12 MISS with every MISS on the two naked asymptotes as pre-declared, the two finite-size zero-parameter comparators HIT every row at every band (residual on ln y*/ln h -0.00022 to -0.00002 against C2), K*/pool falls 0.032 -> 0.026 -> 0.020 -> 0.016 -> 0.013 with no bend and a quiet sup, and the band means sit 0.0017 to 0.0037 below both asymptotes, six to twelve times their 0.000291 difference, so which limit the deficit tends to is set by the comparator and not by the data. | [blind-0830-quadpoint-31607.md](history/staging/blind-0830-quadpoint-31607.md) |\n| Z2 | `Q-quadpoint-c1c2-0830` Does the fourth decade's post-hoc, unsealed preference for the omega-form comparator C2 over the exact-Mertens comparator C1 (C1 above the data by 3.46 and 8.83 se at B11 and B12, C2 within 2.82 se) survive a sealed blind test at the fifth decade, Q in (100003, 316243], and what does that say about which mechanism explains the depth law's drift? | ANSWERED | MEASURED, sealed before the run and scored blind on 17702 new anchors, and it decides nothing asymptotic: the sealed D1 row HITs at both B13 and B14 (C1 above the data by 13.33 and 23.82 se, C2 within 1.26 and 1.82 se), so by the sealed rule the omega-at-finite-u crossing of import-rough-anatomy.md §2.1 is CONFIRMED over the exact-Mertens product of u2-engine-depth.md §5 as the depth law's main term at the sampling scale, while both still HIT every ±0.0015 row, the kill rows on C2 all HIT, K*/pool falls 0.013 -> 0.010 -> 0.008 with no bend, and the band means sit 0.0012 to 0.0018 below both asymptotes, so the two constants 0.280438 and 0.280730 stay unseparated by the pre-declared rule and the run separates two mechanisms, not two limits. | [blind-0830-quadpoint-c1c2.md](history/staging/blind-0830-quadpoint-c1c2.md) |\n| Z2 | `Q-quadpoint-prior-art` In the owning conventions, who holds the rough-pair census, the capture identity and the crossing law, and is any of it ours? | ANSWERED | The object and the identity are standard (the dimension-2 twin sifting function plus Buchstab's identity; ADJACENT-STANDARD, no verbatim carrier, no novelty language); only the comparison X(y) < T and the crossing law stay ABSENT-PER-CONVENTION, and the Buchstab correction is 6.6% of the top-band residual, not 2%. | [quadpoint-prior-art.md](history/staging/quadpoint-prior-art.md) |\n| Z2 | `Q-quadpoint-transplant` Does the anchored-cap family transplanted onto the stretch window S_Q certify every anchor, and what does the depth cost do with height? | ANSWERED | Pre-registration for the decade extension, committed ALONE before the segmented producer existed: it fixes the FALL trend line, the registered band readings, the kill rule and a digit-exact calibration gate that must abort before any new figure prints, and states that a falling K*/pool over two decades would still be only a measurement. | [attack-quadpoint-01.md](history/staging/attack-quadpoint-01.md), [attack-quadpoint-02.md](history/staging/attack-quadpoint-02.md), [quadpoint-decade-prereg.md](history/staging/quadpoint-decade-prereg.md) |\n| Z2 | `Q-recon-0828-rough` Does the 2000-2026 literature on y-rough numbers, sifted sets and short intervals hold any theorem that lowers Z2 or G2, meaning any control on the largest gap or the minimum count of a sifted set in EVERY short interval, or any structure of rough numbers a gap bound could use? | ANSWERED | No. Ten angles examined, ten killed for this purpose; the strongest every-interval statement in print for one-class rough numbers is Iwaniec's S(x,y,z) >= (4y/log^2 y)(log(y/z^2) - O(1)) for y >> z^2, uniform in x, giving J(P(z)) << z^2, and it is the linear sieve run exactly at its sifting limit beta_1 = 2 with no slack outside it; no two-class analogue exists in print at any exponent, the corpus's own x^4.2665 is beta_2 by the same mechanism, and every remaining branch of the field (Matomaki-Radziwill, Gorodetsky, Friedlander-Iwaniec Cor 6.28, Gafni-Tao) carries the almost-all quantifier, which a second-moment statement can never convert into a bound on one empty window. | [recon-0828-rough.md](history/staging/recon-0828-rough.md) |\n| Z2 | `Q-recon-0828-sieve` Is there anything in the 2008-2026 sieve literature, or in the parity-breaking literature, that lowers beta_2 = 4.26645 or supplies a consumer for a level theta > 1 on the two-class comb? | ANSWERED | No route. Twenty-one angles examined, nineteen dead on arrival, nothing in print beats 4.26645 at kappa = 2 at any level on four calibrated channels; the two survivors are an owning-convention row and one unresolved hypothesis in a five-page 1990 announcement whose stated requirement beta < 3 excludes beta_2 = 4.26645. | [recon-0828-sieve.md](history/staging/recon-0828-sieve.md) |\n| Z2 | `Q-redteam-0828-quadpoint` Do the four HELD quadpoint notes of 2026-08-22 survive an adversarial pass: the capture identity's proof line by line, the depth law and its asymptote, the sealed prereg scoring, and the prior-art citations at page level? | ANSWERED | The identity is PROVEN once its window convention is written down and holds at every K of 50 anchors under an independent implementation; four load-bearing sentences are wrong (section 4's gloss on Sum capU, v1 section 2's limit, the prior-art note's 2%, and the note's unconditional y* sentences), and none of the four defects opens or closes a route. | [redteam-0828-quadpoint.md](history/staging/redteam-0828-quadpoint.md) |\n| Z2 | `Q-redteam-0830-records` Do attack-0829n-parity-dstar.md (CLOSED, HELD) and attack-0830-record-mechanism.md (PARTIAL, HELD) survive an adversarial pass on independent code, does REFUTED row 95 stand as worded, and is TODO Z5's new title supported? | ANSWERED | Both notes' measurements reproduce on independent code and neither headline is refuted: D* is confirmed EXACTLY, by a BigInt rational simplex written here, at 299 of 376 anchors with 0 contradictions (the other 77 exceed the exact solver's size cap), the old range's 172-figure gate is re-derived exactly at all 43 anchors with 0 disagreements, and an independent 30-wheel sieve to 1e11 reproduces note (b)'s seven decade gap counts, CV^2 and far-tail slopes to every printed digit, with the counts also matching the published pi_2(10^k). Four sentences are wrong or over-stated and are replaced here: REFUTED row 95, note (a)'s ledger, its section 6 and TODO Z5 all attribute the null's 0.025 agreement to the WIDTH law when 0.025 is the ln Q slope difference (the width-slope difference on the same 84 matched anchors is 0.031); note (a) section 3's \"the threshold sits in a gap of at least 0.25 at every anchor\" is scoped in its own producer to the true arm and the null arm's exact gap is 0.1212; P6's HIT rests entirely on the decade-2 collapse that section 5 discounts as a sampler artefact when it discounts kill clause (a); and note (b)'s \"d b_z = 1.0806 +- 0.0014\" carries only the Monte Carlo pairing error, while moving the far-tail slope by 0.02 (against a spread of 0.0174 across six fit windows on the same decade and 0.1196 across decades) moves d b_z from 0.9888 to 1.1720, and continuing the law at slope 1 above the last resolved point still gives 0.7752. REFUTED row 95's CLOSED verdict stands on kill clause (b), which is better powered than the note presents it (paired s.e. 0.0401 against a 0.15 band), and the row needs two clauses reworded; TODO Z5's title is supported for \"measured and not derived\", and the resolved part of the law carries 72 per cent of the residual on its own with the extrapolation carrying the rest. | [redteam-0830-records.md](history/staging/redteam-0830-records.md) |\n| Z2 | `Q-redteam-0830-zone` Do attack-0829n-X-upper.md, blind-0830-quadpoint-31607.md and blind-0830-quadpoint-c1c2.md survive an adversarial re-derivation and an independent re-implementation of their producers, and at what rung does the corpus carry each of their headline claims? | ANSWERED | They survive on the numbers and mostly on the wording. On independent code every band sum, tier ratio, sealed forecast, band statistic and z of all three notes reproduces to the printed digit (X-upper 7.82 to 10.13 and the flat 378100.0; the fourth decade's 76 rows at 64 HIT and 12 MISS with the same twelve MISSes; the fifth decade's class A at both bands), and three attacks failed to break them (the paired-se attack on the D rows, a search for a fitted constant in C1, and gap-selection at the band edges). Four defects stand: the X-upper note's \"s_max 1.81 to 1.96\" is wrong in its own arrow convention, which means B3 to B8, since s_max peaks at B4 and reads 1.85 at B8; the c1c2 note's OLD R2 row at B14 is a boundary call that my 1-in-8 subsample scores the other way; the c1c2 note's custody line \"three tails, one code-sha256\" is a tautology of tailfmt.js and is not evidence of a seal; and the unscored y* check carries a Jensen bias of about +0.44 in y, which is the whole size of the C2 gap it is read as corroborating. Nothing here moves 4.2665 to 2, no rung rises, and all three notes stay HELD. | [redteam-0830-zone.md](history/staging/redteam-0830-zone.md) |\n| Z2 | `Q-redteam-0904-r0-extension` Do the seven load-bearing claims of derive-0904-r0-extension.md survive an adversarial re-derivation against Tao's own text, and do its proposed live-layer changes survive? | ANSWERED | Six of seven survive at their stated rung and one is refuted: Proposition A, the scope slip, row e'' and the equivariance transfer are CONFIRMED (the slip is strengthened, since Tao's own Example 2 is the target note's missing one-coordinate witness), Propositions B and D are WEAKENED (B's stated mechanism proves too much and its proposed insertion drops the qualifier that saves it; D's \"same trigger\" is near-definitional and its causal reading of the exponent band does not hold), and the claim that no corpus artifact certifies the window-count variance at width p'^2 is REFUTED by research/06-variance-theorem.js, which computes exactly that, brute-force verified at two levels; separately, EXEMPT throughout means \"Claim 1's H5 fails\", which is not the parity statement wall-note Face 1 is priced against, and the note's proposed live-layer text carries no H6 conditional. | [redteam-0904-r0-extension.md](history/staging/redteam-0904-r0-extension.md) |\n| Z2 | `Q-roughpair-error` How big is the empirical error of the rough-pair census, and is Z2 dead on the numbers? | PARTIAL | Not dead on the numbers - the error sits below T at every one of 1,206 anchors, its scatter is sub-Poisson, its systematic part is 0.1% at the top band with the favourable sign and its exponent is 0.40 below the twin count's - but the same run moves the wall off the error term onto the depth, where the slack at the crossing is 1.056 and falling and the main term alone is 5.6 times the twin count at the sifting limit. | [attack-roughpair-error.md](history/staging/attack-roughpair-error.md) |\n| Z2 | `Q-roughpair-null` Does the independent-thinning null named in attack-roughpair-error.md's defects list return chi2/df = 1, and is the measured sub-Poisson dispersion of the rough-pair census error real or a normalisation artefact? | ANSWERED | MEASURED, null-side, label (i), no bearing on the certificate: the independent-thinning null is binomial and returns 1 - p (0.8164 to 0.9808), not 1, so attack-roughpair-error.md's \"below 1 everywhere: sub-Poisson\" is WEAKENED and the registered F1 fires at u = 5; the deficit against the mean-matched null survives at pooled z = -5.87, -7.09, -11.39; and a CRT-exact null that only makes each prime's kill count exact in the window predicts 0.33 to 0.56, which the measured value EXCEEDS by 1.26 to 1.40, so D8 is interval-level in form, elementary in content, and not wall-relevant. | [measure-roughpair-null-0829.md](history/staging/measure-roughpair-null-0829.md) |\n| Z4 | `Q-applied-0828-census` Were the census red team's corrections applied to the three HELD notes? | ANSWERED | Every WEAKENED and REFUTED verdict in redteam-0828-census.md is applied to its target note in the present tense; two producer defects and four live-doc corrections stay open and are listed here. | [applied-0828-census.md](history/staging/applied-0828-census.md) |\n| Z4 | `Q-applied-0828-head` Were the head red team's corrections applied to the three HELD notes? | ANSWERED | Applied. Every WEAKENED and REFUTED row of redteam-0828-head.md lands in the note it targets as 42 prose edits, and all three ledger verdict lines now carry the surviving chain; four corrections fall outside the fence and are owed elsewhere, the load-bearing one being destroyer-census-01 §6(b)'s R and its h/R = 1.09 -> 1.03. | [applied-0828-head.md](history/staging/applied-0828-head.md) |\n| Z4 | `Q-applied-0828-live` Which engine and head red-team corrections reached the live layer? | ANSWERED | Nineteen edits across four documents: the glossary's cap_K dimension count with its mandatory sifted-versus-assumed clause plus three new entries (twin opener, head, tail); the survey's one-class density factor, S1's restored pi(q-1) terms and r in N_x hypothesis, the dropped Chen absorption, Theorem C with its emptiness, dimension 1 for q_i < q, the comb-conditioning qualifier and the inverted shallow-band bullet; the certificate engine's equidistribution status cell, the q = T^{o(1)} second hypothesis and the Comb Discrepancy scope cell; and the census's R as the continuum functional with its three comparators; paper/staircase-note.md, TODO.md and QUESTIONS.md are owed and untouched. | [applied-0828-live.md](history/staging/applied-0828-live.md) |\n| Z4 | `Q-destroyer-census` Who destroys each channel pair, what does the zone kill budget certify, and where does the counting certificate die? | ANSWERED | The census reproduces digit-exactly on an independent engine; the destroyer rule is positional, not temporal, the certificate holds at 15 zones and dies at p = 67 where B/C first crosses 1, and no first-mover excess survives the derived null. | [destroyer-census-01.md](history/staging/destroyer-census-01.md) |\n| Z4 | `Q-head-remainder-0830` Does the 5.3 percent remainder that Hardy-Littlewood leaves in the head's endpoint deficit Delta = 2 - beta derive by a route that is not HL: the frozen-sieve first-survivor object, a level-of-distribution correction, or the exact tile-ensemble head? | PARTIAL | No route that is not HL closes it, and the remainder as recorded is not what it was taken to be. The record's Delta_meas = 0.6214 at [1e7,1e8) pools a decade over which the prime density falls ten percent, and the OLS intercept absorbs the gradient (a Simpson-type effect, sign derived): the two half-decades read 0.6848 and 0.6500, the four quarter-decades 0.7203 to 0.6565, and pooled minus the gap-weighted mean of the halves is -0.0378, -0.0455, -0.0524 at the three decades, so the same-sign 5.3 percent miss is the pooling. At half-decade resolution HL's pipeline prices Delta to 1.2 and 0.4 percent at the top two windows, inside one standard error, and the sign of the miss alternates. That agreement is a near-cancellation of two four-percent effects: the pipeline under-predicts its own rotation ensemble by 3.6 to 4.9 percent (pipe/ens 0.951 to 0.964 across y = 313 to 19997, MEASURED by exact CRT sampling; 1.049 to 0.958 at y = 11 to 23 by full enumeration), an ensemble quantity that derives by finite count with no prime input and whose sign is opposite to hl3 section 2's guess; and the anchored window's deficit sits 2 to 5 percent below its gap-scale-matched ensemble (Delta_meas/Delta_ens 0.878 to 0.995 over six half-decades, 0.954 to 0.995 on the four with more than 13,000 gaps), the anchoring part, which is a three-point prime correlation in a short window and is HL-strength by every route named. The head is a slack field and nothing here touches Z2. | [attack-0830-head-remainder.md](history/staging/attack-0830-head-remainder.md) |\n| Z4 | `Q-head-residual` What is the head's residual prime-origin factor h/R = 1.09 -> 1.03, and does it derive? | PARTIAL | h - R decomposes exactly and A_forced derives; h/R -> 1 follows from beta CV^2 being bounded, beta CV^2 -> 2 controlling the RATE and not the limit (under it h - R -> D = -0.95, so +8.4661/ln^2 p cannot be the limit law); R is the continuum functional and the integer-origin null sees R + 1/2 exactly, so h - R = 5.679 depends on the unstated null population, reading 5.179 discrete-uniform and 2.925 coprime-to-30; the mod-30 answer is NO by a valid route, class term -0.0001; Delta = 2 - beta is priced by HL to 5.3 percent with nothing fitted and HL's own 1/ln x term absorbs the miss, so the 5.5 s.e. framing is dropped; the asymmetry-as-mechanism claim and the [0.289, 0.361] bracket are refuted, hl3's 16 percent standing; beta_null is 1.898 for the discrete-consistent null and 2 in the continuum; the 72/28 split reads 81/19 in the other exact order; c = 0.844 is 0.722 on a subset, pinned to 15 percent; Delta_HL ln p/lnln p = 4.02 is a property of the prediction; and the N4 plateau is not OPEN, needing beta CV^2 = 2.71 rising to 6.63. | [head-residual-factor.md](history/staging/head-residual-factor.md), [head-residual-hl3.md](history/staging/head-residual-hl3.md), [head-residual-null.md](history/staging/head-residual-null.md) |\n| Z4 | `Q-redteam-0828-census` Do destroyer-census-01.md, stretch-01.md and records-placement-01.md survive an independent re-derivation of every load-bearing claim, and may they leave HELD? | ANSWERED | All three survive on arithmetic, every decisive number reproducing exactly on an independent engine; one sealed reading (records READ-3) is REFUTED as vacuous and four sentences need correcting before integration. | [redteam-0828-census.md](history/staging/redteam-0828-census.md) |\n| Z4 | `Q-redteam-0828-head` Do the three 2026-08-28 head-residual notes survive an adversarial re-derivation of every load-bearing claim on independent code? | ANSWERED | Mostly yes on the numbers and no on two framings: \"h/R -> 1 iff beta CV^2 -> 2\" is REFUTED (the limit needs only boundedness, the product controls the RATE), and 9 to 59 percent of the quoted h - R is the choice of comparator, since R is the continuum functional and a discrete uniform origin sees R + 1/2 exactly. | [redteam-0828-head.md](history/staging/redteam-0828-head.md) |\n| Z4 | `Q-redteam-0830-slack` Do attack-0830-head-remainder.md, attack-0830-tail-derivation.md and verify-0830-record-defects.md survive an adversarial re-derivation on independent code, and do the five riders the verification put on live notes stand as written? | ANSWERED | Arithmetically they survive: every figure of the three notes that this pass could recompute reproduced to the printed digit on code written from the definitions, 0 assertion failures over 58 assertion call sites (most inside per-level loops), including every figure of the verification's three claims and the half-decade Delta_HL it quoted rather than recomputed. Four sentences do not survive. (1) verify-0830-record-defects.md:316, a replacement queued for head-residual-hl3.md sec.0, says the pooled value sits \"0.038 to 0.052 below the sub-window mean at all three decades\"; measured, the two lower decades run to 0.076 and 0.066, and the same note's own falsifier row says 0.038 to 0.076, so the wrong half is the one queued to land. (2) verify-0830-record-defects.md:379's un-measured \"+0.005\" residual inside the eighths is +0.0004 measured at sixteenths, so the de-pooled top-decade value is converged and the caution can be dropped. (3) the ruling that the class null is \"the matched figure\" for the tail is matched on residue class only: p'^2 is coprime to every q <= p', and the ensemble's rough-class offset is E_rc - R = 5.6714 -> 7.5357 over x = 7..29 against the class null's 5.6000 -> 6.0336, giving t/(R_shell + 7.5357) = 1.0123 against t/classNull = 1.0157. (4) attack-0830-head-remainder.md sec.3's top row is not gap-scale matched: E[g] 244.0 against 235.9, y_match capped at 19997, and on the note's own six ensemble points Delta_ens ~ E[g]^-0.312, so meas/ens reads 0.9640 not 0.9539 and the headline \"0.954 to 0.995\" reads 0.964 to 0.995. Three levels the notes recorded as out of reach are run here: the X2 group at x = 23, the tile identities and conditionals at x = 29, and the ensemble ladder at y = 29 (pipe/ens 0.9535, 40 s, against the head note's \"about fifteen minutes\"). No route opens or closes and nothing here touches Z2. | [redteam-0830-slack.md](history/staging/redteam-0830-slack.md) |\n| Z4 | `Q-tail-derivation-0830` Does the tail's measured law c = 0.7522 ln^2(p'^2) and its surplus derive, and which part of the coefficient is the tile's own (no prime input) against which part needs Hardy-Littlewood? | PARTIAL | PARTIAL, kill clause fired on the coefficient: the ensemble object (the same functional averaged over all p# translates) derives exactly, E_all = R + 5/2 and E_odd = R + 3 with R = Sum g^2/2W, computed exact at x = 7..29 with every conditional the rough-origin candidate needs, and its Mertens constant e^{2gamma}/(8 C2) = 0.6007 is 1/(2 C2) times rho(2) = e^{2gamma}/4 by construction; every route from there to the anchored 0.7574 passes through rho(2)'s limit (the sharp form of Assumption A, algebraically equivalent to HL) and through the zone's gap shape CV^2 -> 1 (an HL statement), so the anchored coefficient is HL in disguise and nothing beyond HL is derived; the tile's own (1 + CV^2)/2 is 0.6306 -> 0.7532 over x = 7..29 and not settled; the record's 0.7522 = HL x 0.9644 x 1.0298 is two non-HL-sized factors cancelling at this height; the rough-origin candidate has an exact ensemble counterpart E_rc/E_cls = 1.0050 -> 1.0523, right sign, wrong size and trend against the anchored 1.0157, not identified; one number derives that the record measured: the class-null offset E_cls - R = 5.600 -> 6.034 exact against the record's 6.05, replacing both the prereg's 0.5 and the note's 7.5; and the corpus's \"R + 1/2\" is the other convention, under the tail's own the constants are 5/2 and 3. | [attack-0830-tail-derivation.md](history/staging/attack-0830-tail-derivation.md) |\n| Z4 | `Q-verify-record-defects-0830` Do the three defect claims of 2026-08-30 (attack-0830-varE-identification.md sec.8 on the doubled X2 column; attack-0830-tail-derivation.md sec.7 on the corpus's \"R + 1/2\"; attack-0830-head-remainder.md sec.1 on the decade-pooled Delta = 0.6214) reproduce on independent code from the definitions, and which of the corrections they list apply? | ANSWERED | Claim 1 CONFIRMED: the corpus delta*X2 column is the positive half of the p \\| h-2 pattern doubled, that pattern's mirror is the p \\| h+2 pattern (W-(-h) = W+(h) exactly, W-(h) != W-(-h) at 9 to 900,679 shifts), the group sum 2X2 reads -0.5603 .. 0.1846 at x = 7..19 against 5.7147 .. 5.3266, and Xmix is overstated 2.305x at x = 19 (6.9x at x = 7); X, X1 untouched. Claim 2 AMENDED: R + 5/2 and R + 3 are exact for the tail convention (all and odd origins, asserted at x = 7..23) and t/(R + 3) = 1.0228, but head-residual-factor.md:70 is the HEAD's forward convention where R + 1/2 and R + 1 are exact, so that correction does not apply; and p'^2 is 1 mod 6, where the tail constant is 5 (t/(R + 5) = 1.0181), with the class null (1.0157) the matched figure, so 1.0228 is one unmatched convention replacing another. Claim 3 CONFIRMED as an artefact, AMENDED on mechanism and on the half-decade reading: 0.6214 pooled against 0.6592, 0.6705, 0.6736 at 2, 4, 8 sub-windows; the between-slope (Simpson) piece the note derives is half of the pooling term (0.0187 of 0.0378 at halves, 0.0261 of 0.0522 at eighths), the other half is the within slope's own fall with height weighted by Var(g); and re-reading the halves from their eighths puts HL BELOW the measurement at 5 of 6 half-decades by 2.5 to 3.4 percent at the top decade, so \"within 1.6 s.e., sign alternating\" is itself a pooled statement. | [verify-0830-record-defects.md](history/staging/verify-0830-record-defects.md) |\n| Z4 | `Q-zone-tail` What is the tail field of the zone (p, p'^2) - its law against ln^2, its worst case, its share of the R0 width, and is it the frozen-sieve last survivor below p'^2 at the rate the head is the frozen sqrt(p)-level first survivor above p? | PARTIAL | The field exists now, 1,225 zones, and nothing in it is derived. The tail is the head's own renewal functional read at height p'^2 rather than p: c_tail = 0.7771 ln^2(p'^2) at [3163,1e4) against the head's 0.6693 ln^2 p on the same zones and the same estimator, both referenced to HL's 0.7574, and the apparent factor of four is the unit ln^2(p'^2) = 4 ln^2 p and nothing else. The band drift does not settle (0.7518, 0.7288, 0.7771; the pre-registered 0.03 step is missed), so this is band composition on two decades and not a measured convergence. The frozen-sieve mirror is 100% at the zone's own level p, but that is the Zone Restriction Lemma restated and not a rate comparable to the head's 96.65%; the non-vacuous mirror is one fold back, where the tail survives at 98.69% with the residual priced exactly (88 extra lpf(n) = p survivors, 18.18% of them biting), and the head's own frozen level sqrt(p) identifies the tail in only 22.86%, refuting this note's own pre-registered [0,5] zones. The renewal surplus t/R = 1.0619 at B4 is NOT RESOLVED, its bootstrap [0.9937, 1.1321] containing 1, and an exhaustive class-matched origin null over 5,962,057 origins gives the same 1.0619; three matched non-square endpoints see no difference at all, sign tests z = 0.14, -0.54, 0.66 with six paired intervals all containing zero, so nothing here is a property of the endpoint being a prime square. R0 is asserted at all 1,225 zones and the shares are 1.82% head, 89.72% Z2, 8.46% tail at B4 with the tail's share FALLING, so filling the tail row moves the decomposition's difficulty nowhere: Z2 was the obstruction before and is more of it after. | [zone-tail-01.md](history/staging/zone-tail-01.md) |\n| Z4 | `Q-zone-tail-02` On the full zonegap-01 range to 1e11, what is the per-zone tail law in both units, is the renewal surplus t/R real at 27,292 zones, and do the head and the tail of a zone have a joint law? | ANSWERED | Resolved DOWN and nothing is derived: zone-tail-01's unresolved t/R = 1.0619 at 782 zones reads 1.0298 on a height-matched comparator at 17,700 zones (bootstrap [1.0156, 1.0443], excluding 1) and 1.0157 against the exactly matched discrete class null (bootstrap [1.0013, 1.0308], clearing 1 by 0.0013), so part of the original figure was the comparator and the pre-registered rule returns MATCH by two parts in ten thousand; c_local = 0.7522 at the top band with bootstrap [0.7410, 0.7630] now CONTAINING HL's 0.7574 and the band drift inside the registered 0.03 per step for the first time but still non-monotone (0.7518, 0.7664, 0.7416, 0.7522), so no convergence is measured; global worst tail/ln^2(p'^2) = 6.840 at p = 15107 with max/mean no longer flat (4.40, 5.36, 8.59, 8.43); Q5 is answered in the negative, head and tail showing no dependence at 27,292 zones (detrended r = -0.0009 against null s.e. 0.0061, permutation p = 0.8920, variance ratio 0.9990, no KS rejection, no chi-square near threshold), so the R0 pieces are a product structure to first and second moments and in the sum's whole law while independence is NOT proven and the simultaneous-extremes regime is untouched; a candidate rough-origin mechanism for the surplus is HEURISTIC and does not fit the head's own excess band by band; the head field's effective sample size is measured at 1,910 distinct a_first among 17,700 top-band zones, so this note's own head interval is priced over draws the field does not hold, a cluster bootstrap over distinct a_first (producer SEC I(a), seed 613009117) measuring the inflation at 4.10 rather than the 3.04 the redundancy alone implies and moving the top-band interval to [0.6848, 0.7692], still inside the registered [0.68, 0.78] (correction originating with redteam-0829-measure-b.md section 1c C15); the scope of that defect is that one interval and not the corpus, since destroyer-census-01.md publishes no bootstrap, head-residual-factor.js bootstraps over gaps, zonegap-01.js has none and zone-tail-01.js bootstraps only tail quantities; ten of ten predictions hit, which is a criticism of the pre-registration; and the R0 shares read 1.40% / 92.80% / 5.81% with the tail's share still falling, so Z2 is more of the obstruction after this pass than before it. | [zone-tail-02-0829.md](history/staging/zone-tail-02-0829.md) |\n| Z5 | `Q-applied-0828-litimports` Were the lit/imports red team's corrections applied to the six HELD notes? | ANSWERED | Applied, 64 substantive edits across the six notes: the Saffari-Vaughan range exponent is 6/11 in both import notes, the fixed-M THEOREM in import-fracparts is struck and the cap-36 MEASURED to PROVEN upgrade withdrawn, row 15 loses its theorem column and row 16 is regraded stale, and the six weakened sentences in the four literature notes now read as the audit requires; four claims are declined with reasons, and nine live documents plus two fenced staging notes are recorded here as owing corrections that this pass did not apply. | [applied-0828-litimports.md](history/staging/applied-0828-litimports.md) |\n| Z5 | `Q-applied-0828-registries` Which lit/imports corrections reached the registries? | ANSWERED | 13 edits across six files, plus one appended changelog section: IMPORT-MAP gains rows 15 (no THEOREM, the column struck), 16 (STALE) and 17 (CLOSURE, two-sided) with its three counts recounted to seventeen rows; SEARCH-CONVENTIONS gains the Kourbatov-b, generalized-Dickman and VC-of-multiples rows with each channel's calibration state; import-map-construction's determinantal cell, natal-cap-36's separate-census sentence, one REFUTED row and one PRIOR-ART block are corrected; GLOSSARY, TODO, QUESTIONS, the five loose hyperuniformity documents and two staging notes are left for their holders and listed here. | [applied-0828-registries.md](history/staging/applied-0828-registries.md) |\n| Z5 | `Q-record-deficit` Does the 6.0 percent record-location deficit survive the corrected null, and what is it? | MIXED (lit-kourbatov-shortfall.md: ANSWERED; record-location-null.md: PARTIAL) | Survives (z = -3.3, five window cuts); Z5's premises on 'location' and on n_eff were wrong; corrected-null formula in §4. | [lit-kourbatov-shortfall.md](history/staging/lit-kourbatov-shortfall.md), [record-location-null.md](history/staging/record-location-null.md) |\n| Z5 | `Q-record-mechanism-0830` Which of three candidate mechanisms (sub-Poisson gap dispersion, the tile's exact gap law, Kourbatov's k = 1 conspiracy heuristic transported to k = 2) carries the residual 0.78 to 0.97 of Kourbatov's b, at what derived size, and what fraction of b is left? | PARTIAL | The residual is carried by the twin-prime gap law AT HEIGHT and by nothing else tried: a new sieve to 1e11 measures that law under-dispersed (CV^2 rising 0.7276 to 0.9295 over seven decades) with a far tail steeper than exponential (log-slope 1.0638 at the top decade), and fed into the record null it carries 95.7 to 99.8 percent of the residual (d b_z = 1.0370 and 1.0806 +- 0.0014 against 1.0833, MEASURED on a model, paired, 2000 replicates); the tile's period-wide law (d b_z = 8.10) and Kourbatov-Wolf's k = 1 bin-count heuristic (4.20) overshoot by 4 to 7x and are closed as posed; the transport of a bare CV^2 is family-dependent (0.62 against 1.06 at CV^2 = 0.93), so \"computable from Var/E\" is false as stated; Kourbatov-Wolf's second-order scale a_c/abar tracks the measured CV^2 to within 0.04 at every decade with no parameter and is an ansatz, not a derivation; the shape statistic moves a third of the way toward the data (z_D from -1.64 to -0.97); what is left with no derivation is the gap law itself, and the top height band's b (0.937 measured against 1.69 to 1.98 predicted on 17 records) is the open falsifier. | [attack-0830-record-mechanism.md](history/staging/attack-0830-record-mechanism.md) |\n| Z5 | `Q-record-null2` How much of Kourbatov's shortfall coefficient b does an inhomogeneous-intensity record null carry, how much does a 6Z-latticed null carry, and what is the residual? | ANSWERED | The inhomogeneous null carries essentially none of b (d b_z = -0.0012 +- 0.0001, MEASURED); the 6Z lattice's share is LAW-DEPENDENT, +0.1113 +- 0.0003 under the memoryless rounding-up law against -0.0060 +- 0.0003 under rounding to nearest, so the null side is a bracket of 16.1 to 25.1 percent of b and three quarters or more stays a residual with no mechanism; record-location-null.md section 8's argued 1.7e-4 lattice bound is wrong but no single number replaces it; the 15 percent estimator disagreement is pinned as an exact weighting identity; and the shape assumption behind every b reading is measured at 1.6 sigma and stays open. | [measure-record-null2-0829.md](history/staging/measure-record-null2-0829.md) |\n| Z5 | `Q-redteam-0828-litimports` Do the six notes of 2026-08-28 survive an adversarial pass, and which of their load-bearing claims are wrong? | ANSWERED | Every measured number reproduces on independent code; Saffari-Vaughan Theorem 10's range exponent is 6/11 not 1/11, which voids import-fracparts's one banked THEOREM at fixed M and corrects row 15 twice; four further weakenings, one refuted verdict line, no note refuted whole. | [redteam-0828-litimports.md](history/staging/redteam-0828-litimports.md) |\n| Z5 | `Q-redteam-0830-records` Do attack-0829n-parity-dstar.md (CLOSED, HELD) and attack-0830-record-mechanism.md (PARTIAL, HELD) survive an adversarial pass on independent code, does REFUTED row 95 stand as worded, and is TODO Z5's new title supported? | ANSWERED | Both notes' measurements reproduce on independent code and neither headline is refuted: D* is confirmed EXACTLY, by a BigInt rational simplex written here, at 299 of 376 anchors with 0 contradictions (the other 77 exceed the exact solver's size cap), the old range's 172-figure gate is re-derived exactly at all 43 anchors with 0 disagreements, and an independent 30-wheel sieve to 1e11 reproduces note (b)'s seven decade gap counts, CV^2 and far-tail slopes to every printed digit, with the counts also matching the published pi_2(10^k). Four sentences are wrong or over-stated and are replaced here: REFUTED row 95, note (a)'s ledger, its section 6 and TODO Z5 all attribute the null's 0.025 agreement to the WIDTH law when 0.025 is the ln Q slope difference (the width-slope difference on the same 84 matched anchors is 0.031); note (a) section 3's \"the threshold sits in a gap of at least 0.25 at every anchor\" is scoped in its own producer to the true arm and the null arm's exact gap is 0.1212; P6's HIT rests entirely on the decade-2 collapse that section 5 discounts as a sampler artefact when it discounts kill clause (a); and note (b)'s \"d b_z = 1.0806 +- 0.0014\" carries only the Monte Carlo pairing error, while moving the far-tail slope by 0.02 (against a spread of 0.0174 across six fit windows on the same decade and 0.1196 across decades) moves d b_z from 0.9888 to 1.1720, and continuing the law at slope 1 above the last resolved point still gives 0.7752. REFUTED row 95's CLOSED verdict stands on kill clause (b), which is better powered than the note presents it (paired s.e. 0.0401 against a 0.15 band), and the row needs two clauses reworded; TODO Z5's title is supported for \"measured and not derived\", and the resolved part of the law carries 72 per cent of the residual on its own with the extrapolation carrying the rest. | [redteam-0830-records.md](history/staging/redteam-0830-records.md) |\n| D | `Q-doubling-C2` Does the doubling inequality hold on the base-2 chain, and can a bridging certificate prove it? | PARTIAL | Not proven and not refuted: the exact C2 table has sup 5.2727 at s = 16, eleven proven finite-level bounds C2 <= K*+1 tight to a factor <= 2.91, two closures refuted outright, and the alarm is that K* drifts up (slope 0.6881 +/- 0.1328) while C2 does not (0.2018 +/- 0.1412). | [attack-doubling-01.md](history/staging/attack-doubling-01.md) |\n| D | `Q-doubling-bridge-0829n` Can a bridging certificate carry Ghat(2s) <= 8 Ghat(s) from level s to level 2s uniformly in s, all s, on the base-2 chain? | PARTIAL | Not by any proven mechanism: the K*-product bridge is CLOSED at every C2 by the cited run floor K* >= pi(2s)-pi(s) (K* = 17 at s = 16 by exact walk, VERIFIED, so the certificate reads 18 against 8 on the chain itself; it exits the whole legal band at s = 128, PROVEN); the sharper maxsum bridge Ghat(2s) <= maxsum_{K*+1}(T_s) is PROVEN and holds under 8 at all fourteen enumerable steps (VERIFIED), but its all-s form needs an upper bound on K* against the entering primes' two-class covering that nothing proven supplies; the doubling inequality itself is untouched. | [attack-0829n-doubling-bridge.md](history/staging/attack-0829n-doubling-bridge.md) |\n| D | `Q-doubling-killrun-0830` Can the exact per-fold L bounds of the pi(2s) - pi(s) folds inside one doubling step compose to an upper bound on the weighted kill-run below the allowance 8 Ghat(s)/gbar(s), for all s, without passing through K*? | PARTIAL | No, and the composition is closed as a route, not merely unproven: the folds compose as a PRODUCT, K*+1 <= prod_j (1+L_j) (PROVEN here, exact at fourteen steps, 540 against 18 at s = 16), the composed certificate never sits below yesterday's maxsum certificate (PROVEN by monotonicity), it exceeds the allowance at 8 of 14 enumerable steps starting at s = 9 (VERIFIED), and since every fold kills a slot the composed index is at least 2^N, whose floor 2^N gbar(s) alone exceeds 8 Ghat(s) at the chain rungs 16, 32, 64 and beats the cited polynomial ceiling on Ghat for all large s (PROVEN given the ceiling); the sum form is false at seven steps and the max form is a floor; the weighted run itself, computed exactly, sits at 0.83 of its allowance at s = 16 and (M8) is exactly where yesterday left it, OPEN. | [attack-0830-doubling-killrun.md](history/staging/attack-0830-doubling-killrun.md) |\n| D | `Q-kstar-prereg` What is K* at the three next doubling steps, predicted before any period walk? | OPEN | Pre-registration only, committed alone: the predictions, the scoring rule and the growth-type verdict thresholds are fixed in advance, with the inclusion-exclusion engine validated against an independent scan engine on all eleven known steps first. | [attack-kstar-01-prereg.md](history/staging/attack-kstar-01-prereg.md) |\n| D | `Q-recon-0830-rec-killrun` Does the literature hold, in its own conventions, a theorem whose hypotheses REC(s, u0) (the sup-versus-rms recovery of the level-D signed remainder over all positions, attack-0829n-rml-proof.md section 3) or the weighted kill-run K*(s) (the longest run of level-s slots the primes in (s, 2s] can kill, attack-0829n-doubling-bridge.md sections 0 and 3) satisfies as stated, or a Maier-type theorem that makes REC false for two-class sifted sets at some u0 below beta_2? | ANSWERED | No theorem applies to either object as stated, on four calibrated channels searched in the owning conventions; every neighbour is graded NEAREST with its unmet hypothesis named, and neither object is on any refuted row. On the Maier question the answer is negative for REC as stated (the Maier family bounds the COUNT, i.e. the full-level remainder, never a level-D truncation) but carries one calibration: at one class and full level the REC-shaped inequality is FALSE asymptotically, by Buchstab's origin ratio against the Montgomery-Vaughan full-period variance ceiling, and the embedded arithmetic puts the level where that falsity first shows at log10 z between 16 and 141 depending on u0 and epsilon, so a finite-z margin of the kind the corpus measures cannot see a failure of this type; at two classes the same full-level statement is HL-conditional at u0 = 2 and unlocated in print at any u0. No exponent moved. | [recon-0830-rec-killrun.md](history/staging/recon-0830-rec-killrun.md) |\n| D | `Q-redteam-0830-doubling` Do attack-0829n-doubling-bridge.md and attack-0830-doubling-killrun.md survive an adversarial re-derivation on an engine sharing nothing with their producers, and do REFUTED rows 94 and 98 stand as worded? | ANSWERED | The mathematics survives: the maxsum certificate Ghat(2s) <= maxsum_{K*+1}(T_s) and the product composition K*+1 <= prod(1+L_j) are each re-derived here and hold, the second at 21,641,346 nesting links over 6,012,804 killed runs with zero failures including the zero-kill fold, and every quoted figure reproduces digit for digit on a fresh engine (K*(16) = 17; G2(31#) = 348 @ 8813641451 x4 from both base tiles; sup msc 6.6364; the diagonal cells; the 2^N ratios 1.226/1.481/41.348). Row 94 STANDS and is if anything under-claimed, but its attribution is wrong in one direction: Lemma 1 at s = 128 alone clears the whole band, so the s = 16 walk corroborates rather than carries it. One statement is falsified: the bridge note's NOT-REACHED line puts 19#->37# out of reach, but column-major it returns here, reproducing the ladder row x = 37 and adding a FIFTEENTH step at s = 19, 20 (K* = 13, N = 4, C2 3.5200, certificate 3.8000), so the enumerable range ends at s = 20, not s = 18. Row 98 STANDS on its mathematics and is WEAKENED on one clause: \"the truth sits under it everywhere (sup w/a = 0.8295)\" attaches the CERTIFICATE's ratio to the word truth; the truth's sup is 0.6591. | [redteam-0830-doubling.md](history/staging/redteam-0830-doubling.md) |\n| D | `Q-redteam-0830-imports` Do the six lighter notes of 2026-08-30 (varE-identification, smooth-aps, rec-killrun, skeleton-door, at43-bigint, coherence) survive an adversarial pass on independent code and at the source pages, and which of their load-bearing claims are wrong? | ANSWERED | No note is refuted whole and no exponent moves. Every decisive measurement reproduces on code sharing nothing with the producers: the varE mixed remainder and its 0.844 / 0.826 share below 2L and 94.4 / 98.7 percent unbalanced share at x = 13, 17; the Buchstab deviations and all three crossing levels log10 z* = 33.53, 68.80, 33.62; the skeleton door's 9.2 / -0.9 percent, G30_agg 0.1113 / 0.1011 and the 0.0102 cap at @13, @17, by a path that also verifies cap-36 Theorem A; the @43 parity exactness on all 26,157,448 scour classes with 0 above 2^54 and a constructed @47 counterexample; and the five recon-0828 verdict counts summing to 75, which REFUTES rec-killrun's brief-error clause and confirms the orchestrator's rider. Seven claims WEAKEN. The load-bearing one is smooth-aps's \"the weight is unmet in print\": Harper's J. London Math. Soc. 112 (2025) e70293 proves a Barban-Davenport-Halberstam asymptotic for an ARBITRARY sequence on exactly that dyadic all-classes object, and his own footnote says the missing max over x' can be incorporated, so the NEAREST citation changes even though the step still does not close (the 2025 range is sqrt(2x) < Q <= x and its Theorem 2 route bars sieved sets by name). Also: Henriot Corollary 2 applies, but to the ERRATUM's statement and under two hypotheses the note leaves unstated (eps < alpha/600, not \"every eps\"; F at prime powers); varE section 4's display (*) carries an unfilled placeholder and is vacuous by a factor L/d; the skeleton door's \"would tighten\" is wrong in sign at @17, where the closable block is negative and removing it RAISES G30_agg to 0.1020; Halasz-Montgomery is not applicable for the reason given (it is an abstract inner-product inequality); FGKT's heuristic quote is on p. 4, not p. 3; and the \"~13x had no source\" is a corpus miss, the retired G2-walk figure at CHANGELOG.md:5716. Not TPC-strength; nothing here opens a route. | [redteam-0830-imports.md](history/staging/redteam-0830-imports.md) |\n| 0 | `Q-applied-0828-closures` Were the closures red team's corrections applied to the five HELD notes? | ANSWERED | Applied. 47 prose corrections across the five notes; c2prime-refit-22 drops from CLOSED to PARTIAL and three ledger verdicts are rewritten; four producer READINGS blocks and eleven live-document lines carry the same defects and are listed here uncorrected, behind the fence. | [applied-0828-closures.md](history/staging/applied-0828-closures.md) |\n| 0 | `Q-barrier-kappa2` Does Granville's exceptional-character construction, which makes the Jurkat-Richert f and F attained by interval sieving problems at kappa = 1, transfer to kappa = 2 and supply the missing barrier on wall-note Face 4, so that no class-blind argument can beat beta_2 = 4.26645 on intervals? | CLOSED | No. Verdict (b): the transfer runs, fixes the axiom obstruction that sift-limit-attack.md section 2 blamed for the mixed-sign sketch's death, and then hits a hard ceiling at u = 2, because the character's only gift is the one-point identity lambda = chi on z-rough integers and a sign pattern with a +1 coordinate is empty exactly when z^2 >= x; above u = 2 the surviving question is the shifted intersection of two positive-density lambda-classes, which is Chowla-strength, so the section 2 verdict stands with a corrected reason; the band (2, 4.26645] is untouched, and the hypothesis itself implies the target by Heath-Brown 1983. | [attack-barrier-kappa2.md](history/staging/attack-barrier-kappa2.md) |\n| 0 | `Q-derive-0904-L7-transfer` Is the legal open-set target L7, G2(x#) << g(x#) (ln x)^A for a fixed A (which would give exponent 2 + o(1) from Iwaniec's one-class bound, below beta2 and above the TPC line), reachable by a TRANSFER from the one-class Jacobsthal bound, and if not, what does each transfer mechanism reduce to? | PARTIAL | No transfer mechanism in the corpus reaches L7, DERIVED: the union-bound transfer is vacuous already in the period average from x = 11 (the exact CRT budget sum of 1/(p - 1) over 3 <= p <= x passes 1 at x = 11), and the sieve-on-holes transfer is the Bruedern-Fouvry vector sieve, whose coupled unconditional form lands at 2(1 + sqrt e) = 5.2974 above beta2 and whose decoupled form needs the sup-over-positions remainder law that item 0 ruled a truth gap; L7 stays a legal OPEN statement with no mechanism, and the conjectural A = 1 is truth-side evidence only. | [derive-0904-L7-transfer.md](history/staging/derive-0904-L7-transfer.md) |\n| 0 | `Q-g2-provenance-x37` Does the headline G2 exponent depend on the eight terms this repository has never reproduced, and does any instrument without G2(37#) inside it flag x = 37? | ANSWERED | MEASURED, both largely negative: the custody-only refit lands at 1.533 against the headline 1.498 and no provenance-block drop moves the corrected central by more than 0.035 (Ghat(64) alone moves it 0.002), while no G2-free instrument flags 37 in the high direction, the one \\|z\\| past 2 being h(37#) at -2.20 in the low direction and range-dependent; the blind extreme-value forecast of G2(37#) does overshoot at z = +6.58, unexplained. | [measure-g2-provenance-0829.md](history/staging/measure-g2-provenance-0829.md) |\n| 0 | `Q-gap-spectrum` What is the full gap-length distribution of twin slots over one period of the tile, what does the excess functional \\|B_N\\| = sum_{g>N}(g-N) read on an x^s grid, and where does the empirical minimum window count leave zero? | ANSWERED | The distribution is exact at nine levels to x = 31 and the pre-registered renewal null FAILS in the light direction at every one: the tail is uniformly steeper than exponential (body slope -1.3302 against -1), no power law survives (the power-law slope moves by a factor 17), the maximum is 2.087 to 2.241 BELOW the renewal maximum, and \\|B_N\\| sits at 0.0245 to 0.7436 of the null and is exactly zero at s = 1.75 where the null predicts 6.5262e+5 empty windows; so no heavy-tail mechanism for a super-linear G2 exponent exists at any computable level, the identity G2 <= N + \\|B_N\\| is proven but degenerates to G2 <= N on this grid, and the sifting onset is flat at 1.63 to 1.70 against a sieve bound f2 that is zero everywhere the grid reaches. | [gap-spectrum-01.md](history/staging/gap-spectrum-01.md) |\n| 0 | `Q-lit-siebert` Is Siebert's 1983 result stated for general sieve dimension kappa, or for kappa = 1 only, and does it therefore already contain the kappa = 2 extremal example that SEARCH-CONVENTIONS section 3 records as not found? | ANSWERED | kappa = 1 ONLY, on three independent statements of the theorem (the chapter's own publisher abstract, Granville 2020 section 1, Friedlander-Iwaniec 2022 page 20), all naming the LINEAR sieve and the Jurkat-Richert f and F; the chapter's full text was NOT reached, so this is testimony not primary reading, and the SEARCH-CONVENTIONS section 3 \"no published kappa = 2 extremal example\" row needs no correction. | [lit-siebert-1983.md](history/staging/lit-siebert-1983.md) |\n| 0 | `Q-maximal-law-tpc` Is the SIEVE Gaussian maximal law TPC-implying? | ANSWERED | YES in the only form that would deliver anything: stated for the operative remainder R_H, need/z^2 runs 0.49 to 0.61, flat and below the zone budget at every z from 13 to 43, which is the Gap Reformulation, hence the Zone Postulate, hence TPC, so no soft argument can prove it; the published above-2 theta column is not the maximal law's verdict but the law plus two lossy steps. | [phase1-T4-maximal-law.md](history/staging/phase1-T4-maximal-law.md) |\n| 0 | `Q-measure-g2-generic` Where does G2(x#) sit in the distribution of the same maximum gap taken over other two-class configurations with the same census, and is the constant-shift family generic inside the free-pair family? | ANSWERED | G2 is GENERIC in this coordinate on the registered rule, MEASURED: the midrank percentile of G2 in the census-matched two-class ensemble reads 20.00, 18.06, 26.35, 17.14 exhaustively at x = 11..19 and 13.70 sampled at x = 23, all inside the registered [5, 95] band so the falsifier does not fire, while sitting about one ensemble sd below the mean at all five levels; and the free-pair and constant-shift ensembles are PROVEN identical up to translation, so killer 1 gains no target in the class positions at these levels. | [measure-g2-generic-0829.md](history/staging/measure-g2-generic-0829.md) |\n| 0 | `Q-omega-floor-blind-0830` Does the local log-slope of the pointwise floor Omega(z) of attack-0830-rec-cheapest.md (exact to z = 73, certified 2+4-chain family to z = 5e5), extended blind to new levels under a sealed forecast, behave as the DERIVED growth law Omega >> z^{16s/9}/ln^8 z predicts at s = 2.698721, and at what z would the test be decisive? | ANSWERED | MEASURED, blind, on the derived law only: 18 of 22 sealed rows HIT; the certified 2+4-chain family at p* = (47, 43) runs 5e5 -> 1e9 (A1A2 = 8.863e17 -> 9.351e32, exact integers) with its step slope falling 4.85 -> 4.38 through every sealed band onto the law's log-corrected slope, the decisive decade 1e8 -> 1e9 reading 4.410 against 16s/9 - 8/ln z = 4.389 (HIT, kill rule not triggered at 4.452 +/- 0.011); the 4 MISSes are the ratio to z^{16s/9}/ln^8 z sitting at 2.00-2.14 above the sealed cap of 2 at z >= 1e8, the pre-declared non-adverse direction; the exact floor is extended 73 -> 113 (251 -> 1980, equal to the hill-climb floors at 89 and 101), where the law predicts nothing; no exponent moves and the law stays DERIVED and HELD. | [blind-0830-omega-floor.md](history/staging/blind-0830-omega-floor.md) |\n| 0 | `Q-rec-cheapest-0830` At the cheapest legal point of REC(s, u0), s -> 1+sqrt(e) and u0 -> beta_2, where the saving asked over the triangle inequality is only z^{1.031}, does any of the five dead routes close, does a mixed mean-square-plus-cap strategy close, and is the failure a proof gap or a truth gap? | PARTIAL | No route closes and the failing step is yesterday's statement, so PARTIAL on the routes; but the step is located as a TRUTH gap, not a proof gap, asymptotically: the certificate's own value at a CRT-planted doubly-smooth window is cc(r) = -(A1A2 + A1B2 + B1A2) with every term a count of exit chains of the Rosser supports (PROVEN, checked exhaustively at z <= 73), the window bound T <= cc(r) + H - 1 is exact, and the exit-chain count is DERIVED (held) to grow like z^{16s/9 - o(1)} for s <= 3, the hypothesis the 2026-08-30 red team added to the threshold, and z^{2s - o(1)} in the limit, so REC(s, u0) is false for every u0 < 16s/9 = 4.7088 at s -> 1+sqrt(e), the whole legal band (2, beta_2] included, and RML(alpha) with it, its falsity beyond z ~ 5.6e31; at every computable z it is a proof gap (Omega(z) exact to z = 73 reads 251, and the ledger's former comparator 1.05e6 is HM(u0), not z^{u0} = 7.188e7 at z = 73, so the gap is 68.3x larger than the ledger stated, while a certified lower bound 8.86e17 at z = 5e5 sits a factor 1.2e6 under z^{u0}); no exponent moved. | [attack-0830-rec-cheapest.md](history/staging/attack-0830-rec-cheapest.md) |\n| 0 | `Q-recon-0828-covering` Does the interval-covering, automated-proof or extremal-combinatorics literature hold a method that bounds the length of an interval covered by two residue classes per prime, or that turns the exact G₂ ladder into a statement at all x? | ANSWERED | No route. Fifteen angles examined, fourteen dead on arrival or already closed here; the best interval-covering bound in print is Crittenden-Vanden Eynden's own Lemma 2, which reads 1.02e12 against G₂(79#) = 1710 and is 8181x weaker than the corpus's own 4.2665 sieve bound at the same level; the one survivor, transplanting Costello-Watts's recurrent certificate to two classes, is a computation that certifies finitely many levels and whose analytic half is Erdős #970 at one class. | [recon-0828-covering.md](history/staging/recon-0828-covering.md) |\n| 0 | `Q-recon-0828-farfields` Do any of the fields still absent from IMPORT-MAP.md hold a theorem whose hypothesis the tile satisfies exactly and whose conclusion is a lower bound on a periodic set's count in every window, a maximal-gap bound for a product-structured periodic set, or a positivity statement at a distinguished point? | ANSWERED | No route from any of the sixteen candidates priced; two reach STRONG-ANALOGY or better and CLEAN and both bank only a closure or a wall address, the Fourier one by a one-line non-vanishing proof that empties the Donoho-Stark/Tao/Meshulam family on this object, and Janson's lower tail by saturating at exp(-1/c) >= 0.0805 against the 1/W a position union bound needs, short by 3.6e2 at x = 11 and 2.8e13 at x = 41. | [recon-0828-farfields.md](history/staging/recon-0828-farfields.md) |\n| 0 | `Q-recon-0828-rough` Does the 2000-2026 literature on y-rough numbers, sifted sets and short intervals hold any theorem that lowers Z2 or G2, meaning any control on the largest gap or the minimum count of a sifted set in EVERY short interval, or any structure of rough numbers a gap bound could use? | ANSWERED | No. Ten angles examined, ten killed for this purpose; the strongest every-interval statement in print for one-class rough numbers is Iwaniec's S(x,y,z) >= (4y/log^2 y)(log(y/z^2) - O(1)) for y >> z^2, uniform in x, giving J(P(z)) << z^2, and it is the linear sieve run exactly at its sifting limit beta_1 = 2 with no slack outside it; no two-class analogue exists in print at any exponent, the corpus's own x^4.2665 is beta_2 by the same mechanism, and every remaining branch of the field (Matomaki-Radziwill, Gorodetsky, Friedlander-Iwaniec Cor 6.28, Gafni-Tao) carries the almost-all quantifier, which a second-moment statement can never convert into a bound on one empty window. | [recon-0828-rough.md](history/staging/recon-0828-rough.md) |\n| 0 | `Q-recon-0828-sieve` Is there anything in the 2008-2026 sieve literature, or in the parity-breaking literature, that lowers beta_2 = 4.26645 or supplies a consumer for a level theta > 1 on the two-class comb? | ANSWERED | No route. Twenty-one angles examined, nineteen dead on arrival, nothing in print beats 4.26645 at kappa = 2 at any level on four calibrated channels; the two survivors are an owning-convention row and one unresolved hypothesis in a five-page 1990 announcement whose stated requirement beta < 3 excludes beta_2 = 4.26645. | [recon-0828-sieve.md](history/staging/recon-0828-sieve.md) |\n| 0 | `Q-recon-0829-escapes` Each of the three killers is an obstruction shape that occurs elsewhere in mathematics; where has each shape been beaten in print, by what mechanism, under what hypothesis, and does that hypothesis hold for the tile and zone objects? | ANSWERED | Of nineteen escapes read, fourteen fail a hypothesis check (twelve on a fact already in this corpus), four fail some other way and one (the decoupled vector sieve) stays OPEN as the programme's existing road, so no route opens; Granville-Soundararajan's uncertainty principle is NOT killer 3 in print, is adverse-direction, and is already cited here twice. | [recon-0829-escapes.md](history/staging/recon-0829-escapes.md) |\n| 0 | `Q-recon-0829-farfields2` Do any of eight fields still absent from IMPORT-MAP.md and from the three 2026-08-28 recon notes hold a theorem whose hypothesis the tile or the zone satisfies exactly and whose conclusion is a lower bound on a periodic set's count in every window, a maximal-gap bound for a product-structured periodic set, or a positivity statement at a distinguished point? | ANSWERED | No route from any of the eight; the pass banks one PUBLISHED-ANCHOR (Holt and Rudd's transfer matrix has its full spectrum in print, second eigenvalue a2 = prod_{17<=q<=p}(q-3)/(q-2), printed 0.10206751799779 at p = 999999999989, decaying only like 2.82/ln p, and this corpus's prior-art record names the eigenstructure and the binomial eigenvectors but carries the per-prime eigenvalue (p-j-1)/(p-2) at lit-pdf-holt-rudd.md line 259 but no product-over-primes closed form, no numerical value and no rate), one CLOSURE with mechanism (Beurling-Selberg is the proof device behind the large sieve, whose constant N-1+1/delta is an l2-to-l2 statement and whose error 1/delta exceeds the window at the operative theta_total > 2, so the whole extremal-function family reproduces a row already rejected), and three wall addresses; two angles are TPC-strength at the form that would pay and are labelled (ii). | [recon-0829-farfields2.md](history/staging/recon-0829-farfields2.md) |\n| 0 | `Q-recon-0830-rec-killrun` Does the literature hold, in its own conventions, a theorem whose hypotheses REC(s, u0) (the sup-versus-rms recovery of the level-D signed remainder over all positions, attack-0829n-rml-proof.md section 3) or the weighted kill-run K*(s) (the longest run of level-s slots the primes in (s, 2s] can kill, attack-0829n-doubling-bridge.md sections 0 and 3) satisfies as stated, or a Maier-type theorem that makes REC false for two-class sifted sets at some u0 below beta_2? | ANSWERED | No theorem applies to either object as stated, on four calibrated channels searched in the owning conventions; every neighbour is graded NEAREST with its unmet hypothesis named, and neither object is on any refuted row. On the Maier question the answer is negative for REC as stated (the Maier family bounds the COUNT, i.e. the full-level remainder, never a level-D truncation) but carries one calibration: at one class and full level the REC-shaped inequality is FALSE asymptotically, by Buchstab's origin ratio against the Montgomery-Vaughan full-period variance ceiling, and the embedded arithmetic puts the level where that falsity first shows at log10 z between 16 and 141 depending on u0 and epsilon, so a finite-z margin of the kind the corpus measures cannot see a failure of this type; at two classes the same full-level statement is HL-conditional at u0 = 2 and unlocated in print at any u0. No exponent moved. | [recon-0830-rec-killrun.md](history/staging/recon-0830-rec-killrun.md) |\n| 0 | `Q-recon-0904-sifting-limit-floor` What is actually known, in the owning convention and at the page, about LOWER bounds on sifting limits at dimension kappa > 1, and is the class-blind cap at beta_2 = 4.26645 a barrier or a method artefact? | PARTIAL | Both. The corpus's [ABSENT] is wrong for lower bounds and right for extremal examples: Selberg's Lectures section 17 and Brady's 2017 Corollaries 1 and 3 ARE lower bounds on beta(kappa) at kappa > 1, invisible because Selberg's convention is the reciprocal a_k = 1/beta_kappa so his lower bounds are titled \"upper bounds for sifting limits\", while no kappa > 1 extremal example exists and Halberstam's own 2003 review says so; the best floor on beta(2) resting on a citation is 1.8394 from Brady's reduction evaluated at kappa = 2, unconditional and below the band, and beta(2) >= 2 follows from monotonicity by a padding argument DERIVED here and probably folklore; the cap at 4.26645 as a METHOD ARTEFACT at rung MEASURED is REFUTED as stated (rider 2026-09-04, from `redteam-0904-sifting-limit.md`, on four independent grounds): the level-D LP fixes one omega-profile out of the class Ford defines by a one-sided inequality, and a legal profile inside the classical budget at x = 13 reads 5.0113 calibrated, above beta_2, firing the note's own pre-registered refutation; the ledger's \"3.3152 sits about 18 per cent below the DHR value\" is 22.3 per cent; and its \"passed, 0 of 15\" external test does not survive, so the calibrated sentence is that no barrier theorem exists, no axiom-only argument inside the band is exhibited, and the LP number is a finite-level measurement for one profile; the band (2, 4.26645] is untouched by anything proven. | [recon-0904-sifting-limit-floor.md](history/staging/recon-0904-sifting-limit-floor.md) |\n| 0 | `Q-redteam-0828-closures` Do the five 2026-08-28 closure notes stand, and do they stand for the reasons they give? | ANSWERED | Three stand on corrected sentences (1d, Z5b, 0); one stands in half (10, the 1.074 counter-claim does not); one does not stand for its stated reason (1c, closed on a narrower window than the prior work's significant one, and its CLOSED has overwritten a PARTIAL in the generated index). | [redteam-0828-closures.md](history/staging/redteam-0828-closures.md) |\n| 0 | `Q-redteam-0830-floor-growth` Does the DERIVED growth half of attack-0830-rec-cheapest.md sec.4.2 (Omega >> z^{16s/9}/ln^8 z from four-prime Rosser exit chains) survive an adversarial re-derivation on independent code, does it really make REC(s, u0) false for every u0 < 16s/9, and does the blind slope test of blind-0830-omega-floor.md test the LAW or the CONSTRUCTION? | ANSWERED | No break found in six attacks, so the growth half SURVIVES this pass and stays DERIVED (one failed break is not a proof, and the whole thing still rests on the exact half E1-E3 that this pass did not grade); but four claims are corrected - the threshold 16s/9 needs the unstated hypothesis s <= 3 (above it the four-prime LP maximum is 2(2s+2)/3, then 8, both below 16s/9, though the corollary that every legal u0 < beta_2 fails survives at every admissible s because 2 x LP-max > beta_2 for all s > 9beta_2/16 = 2.3999), the 2s - o(1) limit needs 3^k <= ln D / ln 8 so only k = 2 is available at every z below 1.1e9, A_i is >= not ~ the dyadic count, and the certified ratio does NOT rise monotonically (0.0118, 0.0128, 0.0101, 0.0057, 0.0051, 0.0093, 0.0336, 0.0801, 0.1658, 0.3545 on my own recount, which reproduces all ten S4b rows digit for digit); the blind test is ruled a test of the CONSTRUCTION's count and not of the law, since it cannot see E1-E3 or the CRT step; on the ruling Chris asked for, the floor does make the RML route a truth gap asymptotically and conditionally on E1-E4, and changes nothing computable, the literal construction first passing H = z^{u0} at z ~ 5.6e31 on the constant 2.93e-4 measured here; no exponent moved. | [redteam-0830-floor-growth.md](history/staging/redteam-0830-floor-growth.md) |\n| 0 | `Q-redteam-0830-floor-sign` Does the EXACT half of attack-0830-rec-cheapest.md sec.4 (E1-E4, the sign facts, the window bound T <= cc(r) + H - 1, the exact Omega, and sec.(d)/(e)'s calibration sentences) survive an adversarial re-derivation on independent code, and does it by itself constrain REC at any u0 without the growth law? | ANSWERED | The mathematics STANDS as PROVEN and is stronger in two places than the note claims, but it constrains REC at NO u0 by itself: E3's identity is pure algebra and holds at every position (0 mismatches in 20.5M positions, both s), E2 holds at every divisor n > 1 under three disjoint routes at every s >= 1 and FAILS at s = 0.8, locating a hypothesis D >= z the note omits, E4 has 0 violations at every x over four complete periods with the wrap, and Omega is reproduced by an independent enumeration at all 26 tabled values and EXTENDED to z = 101 (285, 441, 684, 990, 1155 at z = 79, 83, 89, 97, 101), confirming blind-0830's exact 684 and 1155; three sentences are wrong -- E1's \"every pair with gcd \\| 2 is realised\" (exactly half are; the missing half is all those with 2 dividing one side), the ledger's and sec.0's \"251 against z^{u0} = 1.05e6\" (1.05e6 is HM, z^{u0} = 7.19e7, so the gap is understated 68.3x), and sec.4.4's reason for lam+(P(z)) = 0 (0.24 exceeds 1/8; the figure that closes the slab is 0.1199 at z = 89, under 1/8 by 4 percent) -- and sup\\|rho~\\| >= (Omega - M)/2 is loose: sup\\|R_1\\| = Omega + M exactly. Every exactly known log_z Omega is under 2 while REC needs it above u0 > 2, so the exact half touches REC at no u0 and every adverse consequence in the note still rests entirely on the growth half. No exponent moves. | [redteam-0830-floor-sign.md](history/staging/redteam-0830-floor-sign.md) |\n| 0 | `Q-redteam-0830-imports` Do the six lighter notes of 2026-08-30 (varE-identification, smooth-aps, rec-killrun, skeleton-door, at43-bigint, coherence) survive an adversarial pass on independent code and at the source pages, and which of their load-bearing claims are wrong? | ANSWERED | No note is refuted whole and no exponent moves. Every decisive measurement reproduces on code sharing nothing with the producers: the varE mixed remainder and its 0.844 / 0.826 share below 2L and 94.4 / 98.7 percent unbalanced share at x = 13, 17; the Buchstab deviations and all three crossing levels log10 z* = 33.53, 68.80, 33.62; the skeleton door's 9.2 / -0.9 percent, G30_agg 0.1113 / 0.1011 and the 0.0102 cap at @13, @17, by a path that also verifies cap-36 Theorem A; the @43 parity exactness on all 26,157,448 scour classes with 0 above 2^54 and a constructed @47 counterexample; and the five recon-0828 verdict counts summing to 75, which REFUTES rec-killrun's brief-error clause and confirms the orchestrator's rider. Seven claims WEAKEN. The load-bearing one is smooth-aps's \"the weight is unmet in print\": Harper's J. London Math. Soc. 112 (2025) e70293 proves a Barban-Davenport-Halberstam asymptotic for an ARBITRARY sequence on exactly that dyadic all-classes object, and his own footnote says the missing max over x' can be incorporated, so the NEAREST citation changes even though the step still does not close (the 2025 range is sqrt(2x) < Q <= x and its Theorem 2 route bars sieved sets by name). Also: Henriot Corollary 2 applies, but to the ERRATUM's statement and under two hypotheses the note leaves unstated (eps < alpha/600, not \"every eps\"; F at prime powers); varE section 4's display (*) carries an unfilled placeholder and is vacuous by a factor L/d; the skeleton door's \"would tighten\" is wrong in sign at @17, where the closable block is negative and removing it RAISES G30_agg to 0.1020; Halasz-Montgomery is not applicable for the reason given (it is an abstract inner-product inequality); FGKT's heuristic quote is on p. 4, not p. 3; and the \"~13x had no source\" is a corpus miss, the retired G2-walk figure at CHANGELOG.md:5716. Not TPC-strength; nothing here opens a route. | [redteam-0830-imports.md](history/staging/redteam-0830-imports.md) |\n| 0 | `Q-redteam-0830-rml` Do the three HELD notes of 2026-08-29 night and 2026-08-30 on the RML chain survive an adversarial pass — the six-arrow chain and the term-count cap lemma of attack-0829n-rml-proof.md, the same-object ruling and the applied replacement sentences of verify-0830-usup-convention.md, and the reflection identity, custody, pilot bounds and sealed forecast of measure-0830-rho-sup-z41.md? | ANSWERED | The three load-bearing PROVEN claims survive independent re-derivation — the CAP lemma (every step, sine sum at worst ratio 0.585568, the constant 6, 0 per-modulus violations at z = 13, 17), the reflection identity rho~(W-3-y) = -rho~(y) (proved here and its two hypotheses verified at z = 13..37, residual <= 1.7e-10 at every position of z = 13, 17, 19), and the same-object ruling (an independent lattice reproduces the repository's term multiset exactly at z = 13..37 and its {Vabs > 0} set equals LSE's records at all ten levels); five of the seven cited exact sups and three of the five walked sup\\|R_H3\\| reproduce to <= 5.0e-7, and every figure of the sealed z = 41 forecast reproduces from the seven cited sups alone. Fourteen claims do not survive, none of them reopening a route. The load-bearing one is RML sec.2's ms_m/ask column: it measures the PROVEN mean-square BOUND sqrt(B H), not the ^{1/2} that REC names, the two differing by 36.53x to 200.45x on the tabled rows because saturates in H while B H does not, so sec.5's ordering (ask below the Gaussian factor at u0 = 3.0 and 3.5) is an artifact and reverses at every row against the true mean square. \"gauss -> 0\" (sec.5) and \"sup/rms <= sqrt(2 lnW) O(1) = z^{o(1)}\" (sec.3) are the same error twice, against the note's own sec.0: sqrt(2 ln W) = z^{1/2+o(1)}, measured gauss 0.7905 at z = 19 falling to 0.5645 at z = 1e7, limit 1/2. \"the recovery exponent is log_z(2 sup\\|rho~\\|/rms): about 0.99-1.29\" quotes the truth column, which is log_z(2 sup\\|rho~\\|) alone; the quantity named measures 0.5628 to 0.6712. \"REC would be moot rather than false\" should read false. \"u_cap -> 2s + o(1)\" and \"n ~ z^{2s}/polylog\" are asserted where only \"<=\" is proven and the data cannot decide: u_cap is 4.2652 at z = 47, crosses beta_2 only at z = 53, reaches 4.4891 at z = 79, and passes u_triv from z = 53. \"seven points spanning less than one octave\" is 1.509 octaves, still uncorrected in three places including the ledger verdict. verify-0830's applied \"polynomially many from z = 223 on\" is true at every z >= 13, and its z = 331 crossover silently sets A = 1 (A fitted to LSE's own Ssat moves it to 337). No exponent moves, no route opens or closes, and no ledger verdict changes. | [redteam-0830-rml.md](history/staging/redteam-0830-rml.md) |\n| 0 | `Q-redteam-0904-floor-growth-2` Does the growth half of the adverse pointwise floor (Omega >> z^{16s/9}/ln^8 z from four-prime Rosser exit chains) survive a SECOND adversarial pass, and does a floor on the certificate at a CRT-planted window make REC(s,u0) false (a truth gap) or only unprovable by that certificate (a proof gap)? | ANSWERED | SURVIVES, at the same rung and no higher: four new attacks and two unasked ones found no break, so the growth half stays DERIVED and red-teamed twice, which is two failed breaks on a derivation and not a proof; the proof-gap reading is REFUTED because REC is a statement about R_H, the remainder of the very certificate the floor bounds below, and the chain has no step where a certificate value must become a true remainder, so it is a truth gap for REC as the corpus states it; five corrections are owed to the record -- the REFUTED row is UNSCOPED and should read \"on the Rosser vector-sieve lattice at equal levels\" since only the Rosser instance is derived, the \"falsity beyond z ~ 5.6e31\" is an s = 2.698721 figure quoted without its s and the fitted crossing at the corpus's own measured family s = 3.0 is z ~ 10^15.3, the analytic p* window (4D^{1/27}, D^{1/9}/4) is neither necessary nor sufficient at reachable levels so the exact onset is 4.42e5 at s = 3.0 rather than the quoted 1.055e6, the four-prime construction has NO exit prime below z once s >= 7 so the first pass's \"8 beyond s = 5\" holds only on [5,7), and the first pass's escape-(a) density 1/W understates the planted class by the factor W/(n1n2) though the escape still fails by an exponential; the LP maximum 8s/9 is confirmed as an exact vertex and the corollary crossing reproduces 9beta2/16 to 4.4e-16, the pre-registered slope band held at a new s (top step 4.9025 at s = 3.0 against a model 4.8990), and no exponent moved. | [redteam-0904-floor-growth-2.md](history/staging/redteam-0904-floor-growth-2.md) |\n| 0 | `Q-redteam-0904-item0` Do the two HELD notes of 2026-09-04 on item 0, the second adversarial pass on the adverse floor's growth half and the L7 one-class-to-two-class transfer, survive an independent adversarial pass, and do the live-layer changes they propose survive verbatim? | ANSWERED | Both notes SURVIVE at their own rungs and neither moves an exponent, but SIX of their proposed live-layer changes need rewording and two of their supporting derivations are wrong; every load-bearing number reproduces on independent code (18 of 18 A1A2 cells digit for digit, the LP maximum 8s/9 at 0 deviations of 16 with residual 4.441e-16, the crossing 2.399878284834 against 9beta2/16 at 1.097e-13, the fitted C = 1.1206e-4 and the s = 3.0 crossing 10^15.30, the onsets 4.390e5 / 1.111e6 / 1.860e6, ln(n1n2) = 7.4993e7, and the L7 CRT counts EQUAL at x = 7, 11, 13), so the truth-gap reading of REC is CONFIRMED and the growth half stays DERIVED and red-teamed twice, not proven; what does not survive is the scoping, the exit-window and the price wording: the proposed clause \"at equal levels D1 = D2 = z^s\" is NARROWER than the derivation supports, since over 1643 admissible upper-and-lower level splits the floor beats the decoupled window at every cell by at least 0.644444, so no level split escapes and the sibling rider \"asymmetric levels give 8s1/9 + 8s2/9 with s1 + s2 pinned\" is REFUTED as written (it holds only while both s_i <= 3); the record's z ~ 5.6e31 is u0-specific as well as s-specific (10^31.68 at u0 = 4.2165 against 10^35.36 at u0 = beta2, same s), so replacing it with a beta2-convention 10^15.30 mixes two conventions; the onset correction is right numerically and wrong in wording, since at the record's own s = 2.698721 the exact onset is 1.875e6, LATER than the quoted analytic 1.055e6, not earlier; the escape-(a) correction is right in substance but its \"(A1A2)^2 / (2 n1n2)\" is a LOWER bound on the split's contribution, and the two-sided version through lambda+(n) <= 2^omega(n) reads 10^{-3.004323e+7} against B ~ 10^{10.1}, still no contradiction; and on the L7 side the union-bound argument and the CRT count are CONFIRMED while the coupled vector-sieve price 2(1 + sqrt e) = 5.297442541400 is the FORCED-equal-level constant, the corpus's live figure being K_BF = 5.158064680330, reproduced here to twelve digits at (a, b) = (2.229949, 2.928115), and the sieve-on-holes is the vector sieve only after the Ford-Halberstam substitution step that sift-limit-attack.md 7b(1b) already owns; the \"no third mechanism excluded\" line is honest but incomplete, since weighted sieves with Chen switching are already priced and closed at sift-limit-attack.md 4.2; scope of death gains one row, L7's mechanism 2, which the growth note's own section 10 calls untouched while its sibling calls it dead; no exponent moved. | [redteam-0904-item0.md](history/staging/redteam-0904-item0.md) |\n| 0 | `Q-redteam-0904-sifting-limit` Do recon-0904-sifting-limit-floor.md's five claims survive adversarial re-derivation at the page, and in particular is the 4.26645 cap a method artefact at rung MEASURED? | ANSWERED | Four of five claims survive with corrections and the headline does not. Claim d is REFUTED as stated on four grounds, each sufficient: the level-D LP fixes one omega-profile out of a class Ford defines by a ONE-SIDED inequality, and inside the classical (Omega_2(kappa,L)) budget a legal profile at x = 13 gives a calibrated reading of 5.0113, above beta_2 = 4.26645, which fires the target note's own pre-registered refutation of its floor; the calibration is a 1.72 multiplicative correction fitted at one point and it misses the second proven point beta(1/2) = 1 by 66 to 82 per cent, two and a half times the 28.7 per cent effect the headline rests on; the pooled 3.3152 averages four statistics of which only s* is the definitional analogue of beta, and that route reads 3.4678 here and 3.9487 at x = 43 in the corpus's own data, still climbing at +0.4297 per ln x with the fitted line reaching 4.26645 at x = 104; and 3.3152 is 22.2961 per cent below 4.26645 rather than the 18 per cent the note prints. Claim a is CONFIRMED with a second source off the OCR channel (Ford 2023 p. 37, \"Some authors, e.g. Selberg, refer to 1/beta(kappa) as the sieve limit\"), one page correction owed (Halberstam is p. 117, not p. 116). Claim b is CONFIRMED and strengthened: Brady's class is stated on his p. 10 and IS the axiom-only interval class Face 4 needs, his v_R table is reproduced here, and the identity v_{2d+2} = v_{2d+1} his proof asserts is confirmed at d = 0, 1, 2, so 1.819592 rests on proved facts alone. Claim c is CONFIRMED but demoted: beta(2) >= beta(1) = 2 is two lines from Ford's own one-sided (Omega) plus his own beta(1) >= 2, so the padding construction is correct and unnecessary. Of the eight proposed live-layer changes, two survive verbatim, five need rewording and one is refuted; the new SEARCH-CONVENTIONS row adds exactly three clearing phrases and changes nothing for any existing absence claim. | [redteam-0904-sifting-limit.md](history/staging/redteam-0904-sifting-limit.md) |\n| 0 | `Q-rho-maximal-law` What is the rho maximal law, what is its weakest sufficient form against the crossing, and where does it sit against print? | PARTIAL | The pricing lemma turns any RML(alpha) with alpha < beta_2 = 4.26645 into G2(z#) << z^alpha ln^2 z, every step proven except the law itself; the sufficiency curve lambda_max falls from 2.31949 at z = 13 to 1.78446 at z = 47, the weakest measured point, and whether it ever reaches zero is model-dependent and undecided, so nothing here is an unconditional exponent. | [rho-exact-z31-01.md](history/staging/rho-exact-z31-01.md), [rho-maximal-law.md](history/staging/rho-maximal-law.md) |\n| 0 | `Q-rho-sup-z41-0830` What is the exact sup over all positions of the centred sawtooth remainder at z = 41 (period 7.42e12), and of R_H at H = z^3 and z^4, and what does the eighth exact point do to the truth's slope and to the u0 = 3 form of REC(s, u0)? | PARTIAL | Not run in full; the exact z = 41 point prices at 4.8 to 5.4 h on this machine from measured per-position costs (naive 69.8, table 23.5, table-plus-windows 36.6 ns per worker; fleet 4.52 ns of wall at z = 41), above the 3.5 h rule, so what exists is a new engine (a PROVEN reflection that halves the walk, a P(29) periodic table worth 2.9x, per-block exact re-seeding) reproducing every cited exact sup at z = 13..37 digit for digit, sup\\|R_H\\| at H = z^3 walked for the first time at z = 19..37 (F3 = 26.4 to 35.9, the factor 2 was 1.5 to 2.2x loose), a sealed pre-registration (forecast sup 60.07, band [47.25, 76.37], kill needs sup\\|R_H3\\| >= 1807.417), one pilot twelfth giving lower bounds sup\\|rho~\\|(41) >= 70.651250 and sup\\|R_H3\\| >= 64.416936, and an eleven-segment plan for the box; no bound-and-prune exists on three routes; nothing asymptotic can move whatever the point turns out to be. | [measure-0830-rho-sup-z41.md](history/staging/measure-0830-rho-sup-z41.md) |\n| 0 | `Q-rho2-bound` Is there an analytic upper bound on (z) from its proven closed form, and does it matter for RML? | ANSWERED | Already proven 2026-08-21 (attack-rhoms-01); the constant is now named, 170.88 z^{2s} ln^8 z; Chebyshev off any second moment diverges from RML (the true C_L = ms = 1 floor clears at z = 41 and 43 and misses by 2.155x at z = 47), so it is an ingredient, not a route. | [attack-rhoms-01.md](history/staging/attack-rhoms-01.md), [rho2-analytic-bound.md](history/staging/rho2-analytic-bound.md) |\n| 0 | `Q-rml-proof-0829n` Starting from Lemma V's PROVEN mean-square bound, what is the exact chain of implications to a two-class exponent below beta_2 = 4.26645, which single arrow is open, and does it close? | PARTIAL | The chain from the proven mean square to an exponent below beta_2 has exactly one open arrow, the mean-square-to-sup recovery REC(s,u0): sup_x \\|R_H(x)\\| <= z^{u0/2 - eps} rms(R_H) at H = z^{u0} for one fixed u0 in (2, beta_2); every attempt here dies at a named place (term-by-term accounting is capped at 2s + o(1), the trivial-level exponent; position-union pays e^{theta(z)/q}; window smoothing and the P(y) lever move polylog only; the smooth-profile form is closed on hypothesis and priced above beta_2), the arrow is measured TRUE with certified margin z^1.72 to z^1.92 at u0 = 4 over z = 19..37 and the truth's slope 2.766 +/- 0.212 sits 5.8 se under u0 = 4 on seven points under one octave, so the failure is a proof gap at every computable z and undecided asymptotically; no exponent moved. | [attack-0829n-rml-proof.md](history/staging/attack-0829n-rml-proof.md) |\n| 0 | `Q-tail-deficit-model` Does the exact compound-geometric thinning of the previous level's own gap histogram, with the qualifying-gap constraint g = 0, +/-2 (mod q) carried through the ladder, predict the tile's tail-deficit factor G2/(mbar ln D) at zero parameters, and does that make object-models-read-0829's D5 residue-level classification constructive rather than inferred? | ANSWERED | NO. The zero-parameter CRT-constrained thinning predicts a tail HEAVIER than the tile's at every level, delta_M2 = 0.6461, 0.6746, 0.7021 against a measured 0.4577, 0.4463, 0.4791, so the pre-registered \"classification was wrong\" arm fires and D5 returns to OPEN with C3's constructive route REFUTED; the constraint is informative but small, carrying 4.0 to 10.0 percent of ln(null/truth) at the far tail, or 7.7 to 14.4 percent read against a rate-matched control, against the grain's ORDER at 39.4 to 54.6 percent at the deepest abscissa each level resolves; at shallow abscissae the ordering reverses (B 8.7 against C 4.8 at x = 23, u = 4), which the exact operator on a shuffled true grain confirms without any model (maximum 276, 354, 426 against the true 204, 258, 348), and the order effect is now localised (sections 7 and 9): a first-order Markov word from the true adjacent-pair statistics carries 32.3, 34.2 and 60.8 percent of ln(shuffled/truth) at the deepest abscissa of the three folds, and a SECOND-order word from the true adjacent-triple statistics carries 97.3, 97.9 and 102.6 percent there, r = 0.972, 1.061 and 1.000 on draw-mean maxima over 40, 20 and 10 draws, so the folded tile's tail is MEASURED to be a function of the previous grain's three-consecutive-gap statistics and of essentially nothing longer-range; that surrogate is seeded on the true previous grain at one fold, is not iterated, predicts no delta and does not revive C3, and the 348.0 sd 0.0 at 29->31 rests on ten draws. | [measure-tail-deficit-0829.md](history/staging/measure-tail-deficit-0829.md) |\n| 0 | `Q-theta-ladder` Does the all-positions exponent theta turn over below 2, converge to exactly 2, or settle above 2? | CLOSED | The answer is above 2 with an amendment, and the route is retired regardless: theta does not turn over (it dips and comes back, need/z^2 = 0.3550 to 0.6119 at z = 13..31 and rising from 31), convergence to exactly 2 is not supported (e = 1 rejected at 95% in all four fits), point estimates run 2.37 to 2.99 without settling, and OUTCOMES.md records the certificate route as RETIRED because the sharp maximal law it needs is itself TPC-implying. | [theta-ladder.md](theta-ladder.md) |\n| 0 | `Q-usup-convention-0830` Are lemmaV-sup-extension.md's u_sup and u_sat and attack-0829n-rml-proof.md sec.4.1's CAP(z) statements about the same object (level D = z^s at s = 3.0, same moduli, weights and normalisation), so that the sec.4.1 correction in passing applies, or does a level-convention mismatch void it? | ANSWERED | SAME object, VERIFIED at the code: both notes build the lattice through buildTerms(z, z^3), the modulus sets and Vabs(e) agree to 6.9e-18 at z = 13..23 and the count convention matches at all ten levels, the cited Ssat and u_sat reproduce on the RML lattice to every printed digit at z = 13..19 and the full-level alternative does not (19.544 against 19.602 at z = 13, 45.827 against 50.314 at z = 17); so the PROVEN cap Ssat <= CAP <= z^{2s+o(1)} applies and refutes lemmaV-sup-extension.md's asymptotic prose at lines 485-486 and 508-516 (2^pi(z) moduli, a C^pi(z) theorem as the reachable end), while sec.4.1's attribution sentence overreaches by calling that note's \"reading\" a shape it lists as one of two indistinguishable fits and flags as its likeliest error; neither ledger verdict line changes. | [verify-0830-usup-convention.md](history/staging/verify-0830-usup-convention.md) |\n| 1d | `Q-applied-0828-closures` Were the closures red team's corrections applied to the five HELD notes? | ANSWERED | Applied. 47 prose corrections across the five notes; c2prime-refit-22 drops from CLOSED to PARTIAL and three ledger verdicts are rewritten; four producer READINGS blocks and eleven live-document lines carry the same defects and are listed here uncorrected, behind the fence. | [applied-0828-closures.md](history/staging/applied-0828-closures.md) |\n| 1d | `Q-delta-reader-0830` Can any instrument read the sign of the log-power correction delta in G2 ~ c n^beta (ln n)^delta at reach 79, passing the one-class control first? | ANSWERED | No. Six readers were pre-registered against stepped synthetic laws and the one-class control; the four power-chain readers (the diagonal meter among them) are killed by the P(n) stepping alone, the joint fit is calibrated but reads the control's finite-reach delta NEGATIVE (-0.38 +- 0.14 at 64 terms) where its conjectured asymptotic delta is +2, and the one reader that passes does so by importing the exponent, its sign being the sign of (1.777 - beta_hat). delta's sign for G2 is unreadable at reach 79 by any instrument tried, and the control shows a finite-reach reading would not carry the asymptotic sign anyway; item 1d's all-bases form stays unattackable. | [measure-0830-delta-reader.md](history/staging/measure-0830-delta-reader.md) |\n| 1d | `Q-fekete-1d-defect47` Does the bounded-defect Fekete route survive a probe at 47#? | PARTIAL | The defect at 47# is ordinary and the 47# enumeration cannot move item 1d's TPC threshold whatever value it returns; the bounded-defect Fekete lemma is stated exactly and proved with every hypothesis except the candidate itself discharged, so the route reduces to one named inequality and is neither closed nor open beyond that. | [fekete-1d.md](history/staging/fekete-1d.md) |\n| 1d | `Q-g2-falls-decision-rule` Does G2(x#)/x^2 fall, and what pre-registered decision rule would settle it? | PARTIAL | The instrument is calibrated and sharp, resolving the one-class control at 11.6 sigma on the same nine points where it returns 0.9 sigma on G2, and the honest band on the slope is +/-0.117 at nine terms, so the rule is registered; but the power analysis puts separation at x = 53 only if the falling law is x ln^2 x and at x = 151 if it is x ln^3 x, so phase 1's extra terms settle nothing and no reachable exact ladder does either. | [phase1-T3prep-decision-rule.md](history/staging/phase1-T3prep-decision-rule.md) |\n| 1d | `Q-hsub-reductions` Can (H-sub) be proven or refuted, and what does the fold machinery reach? | PARTIAL | Neither, but the statement is smaller: three reductions of the hypothesis are proven, the lemma consuming only power pairs so that it weakens to (H-sub-pow) with the conclusion unchanged, the fold machinery's inability to reach any of them is made exact, and the hunt over all 111 reachable pairs found no drifting family and no counterexample. | [attack-hsub-01.md](history/staging/attack-hsub-01.md) |\n| 1d | `Q-hsubpow-K` Can (H-sub-pow) be proven with an explicit K in the legal zone? | CLOSED | NO by three mechanisms: the CRT lift (K* >= pi(y') - pi(y) diverges), anchored caps (diverge at every base), Iwaniec at two classes (it IS beta2-note); TODO 1d's legal zone is mis-stated, the trusted zone is [1.3946, 11.3568). | [hsubpow-explicit-K.md](history/staging/hsubpow-explicit-K.md) |\n| 1d | `Q-hsubpow-K-0829n` Can (H-sub-pow) be proven with an explicit K inside the trusted legal zone [1.3946, 11.3568) by a mechanism the 2026-08-28 pass did not close? | OPEN | No K is proven at any base; the single open inequality is the uniform-in-k ratio cap G(b^(k+1))/G(b^k) <= e^K G(b), which is a proof gap at a fixed base and a possible truth gap across bases, since for any law G ~ c n^beta (ln n)^delta the all-bases hypothesis holds with finite K if and only if delta >= 0. | [attack-0829n-hsubpow-K.md](history/staging/attack-0829n-hsubpow-K.md) |\n| 1d | `Q-import-interp` Does the Bayati-Gamarnik-Tetali interpolation method reach the G2 Fekete object? | ANSWERED | No: BGT closes on hypothesis H1, since the constraint count never splits (pi(st) - pi(s) - pi(t) is zero at no pair with st >= 25, minimum 2, maximum 12, VERIFIED on 104 pairs), and the one coordinate that does split has limit +infinity; the row lands as WALL-ADDRESS plus the identity that the submultiplicativity defect is the Overshoot slack. | [import-interp.md](history/staging/import-interp.md) |\n| 1d | `Q-redteam-0828-closures` Do the five 2026-08-28 closure notes stand, and do they stand for the reasons they give? | ANSWERED | Three stand on corrected sentences (1d, Z5b, 0); one stands in half (10, the 1.074 counter-claim does not); one does not stand for its stated reason (1c, closed on a narrower window than the prior work's significant one, and its CLOSED has overwritten a PARTIAL in the generated index). | [redteam-0828-closures.md](history/staging/redteam-0828-closures.md) |\n| 1d | `Q-redteam-0830-fekete` Do the load-bearing claims of attack-0829n-hsubpow-K.md (the sign lemma, the K -> beta_bound map, the zone, the TODO.md base-16 correction) and measure-0830-delta-reader.md (the six pre-registered readers, their kill gates, the joint fit's calibration, the RB circularity argument) survive independent re-derivation, and are the two live-doc corrections they caused the right sentences? | ANSWERED | Every number in both notes reproduces on independently written code (all 14 cells of the beta_bound / K_min table, all 48 cells of the reader-calibration table, all 24 figures of the real-control table, the margins, the sups, the 5.607 correction, the 10-of-81 count), and the sign lemma is re-proven as stated for continuous exact laws. Two claims are WEAKENED and one live-doc sentence needs a qualifier: the sign lemma's constant 1.0597 loses 43% of its value if the base floor moves from 2 to 3 and its whole shape fails for the stepped law that a primorial ladder must present, where beta re-enters the sup and a delta = 0 law already needs K = 1.37 at beta = 2, inside the trap; and TODO.md 1d's \"the sign of delta is unreadable for G2 at reach 79 by any instrument tried\" is wrong on the letter, since reader RB passed all four pre-registered criteria and read delta_hat = +0.80 +- 0.09, the note voiding it post-hoc on a circularity the pre-registration flagged but did not make a kill criterion; the honest sentence adds \"non-circular\". No claim in either note is refuted. | [redteam-0830-fekete.md](history/staging/redteam-0830-fekete.md) |\n| 1d | `Q-y2-recompute` Does the sixteen-level Y2 ladder reproduce on the corrected instrument, and does Y2/x^2 fall? | ANSWERED | The ladder reproduces exactly, sixteen of sixteen from sixteen independent processes, digit for digit against the published 2026-08-18 ladder, and Y2/x^2 falls at 15 of the 15 consecutive steps; this does NOT show that G2(x#)/x^2 falls, since the shortfall Y2/(G2-1) is unmeasured above x = 79. | [y2-recompute.md](history/staging/y2-recompute.md) |\n| 0b | `Q-gate-multiplies` Does \"the gate multiplies\" close the entire u-frame recursion branch, or only the merge chain? | ANSWERED | Only the merge chain and one relative of it: the no-fixed-point argument does not reach TODO 0b at all, though 0b is wrong as stated for an unrelated reason and the error is a factor of ln u; what survives is the copy theorem for maxsum (VERIFIED 40 of 40), which needs a residue-deleted maxsum bound that does not pass through a kill count. | [gate-multiplies.md](gate-multiplies.md) |\n| 0b | `Q-import-thinning` Do point-process thinning and interval coalescence give H'' a null model? | ANSWERED | Yes and exactly: the fold recursion is an exactly solvable thinning whose gap pgf transforms by a Mobius map fixing 0 and 1, so under independent thinning the gap word of T_x is exactly geometric with mean mbar/6 at every rung with no error term, which turns both of the record's fitted parameters into derived ones. | [import-thinning.md](history/staging/import-thinning.md) |\n| 0c | `Q-0c-holesweep` At a fold, how does maxsum degrade copy by copy as the two deleted residues sweep through their possible positions? | ANSWERED | The full curve is on disk for nine folds (5->7 through 31->37, every copy) and it is a PALINDROME: the Mirror-Sweep Lemma is proven here and verified exhaustively, giving the fingerprint any kill-count-free bound must reproduce, but no such bound was produced by this pass. | [attack-0c-holesweep.md](history/staging/attack-0c-holesweep.md) |\n| 0c | `Q-0c0e-level-selection` Can a residue-deleted maxsum bound that only has to hold at levels of the argument's choosing carry the weak Zone Postulate? | CLOSED | CLOSED, and the mechanism has a name: the level-selection licence is worth 0.3187 nats once and is falling, no selection rule beats its own null by anything worth having across T_11..T_29, and the BV/EH sweep supplies no moduli-selecting statement to spend. | [attack-0c0e-level-selection.md](history/staging/attack-0c0e-level-selection.md) |\n| 0c | `Q-foldL-maxsum-direct` Can maxsum_m(T_x), plain and residue-deleted, be bounded above by an argument that never counts kills? | CLOSED | The brief's premise contradicts the corpus's own Bridge Floor (maxsum_k >= maxsum_1 = G2(T_x) for every k), so the bridge cannot convert far enough however good the bound is; all four routes close (R1 partial then closed, R2 tautological and quantitative, R3 busts the budget), and the order-m object is in print, so item 0c's absence sentence has to change. | [attack-foldL-05-maxsum-direct.md](history/staging/attack-foldL-05-maxsum-direct.md) |\n| 0c | `Q-import-chaining` Can Dudley's entropy bound or Talagrand's generic chaining beat the union bound over positions on the maxsum law? | CLOSED | No, and the reason is geometric rather than probabilistic: the entropy integral of the true increment metric is already 1.054-1.099 times rms*sqrt(2 lnW) at its lower branch and 1.484-1.545 at its upper, flat across z = 13..23, so chaining's ceiling with a perfect universal constant sits below the union bound's floor; the honest constant-carrying chain measures 2.878 to 3.027. | [import-chaining.md](history/staging/import-chaining.md) |\n| X | `Q-verify-cofactor-convolution` Does the cofactor-convolution identity at the anchor hold, and does it test the joint law? | ANSWERED | RESTATED with corrections: the arithmetic is right, an independent re-derivation reproducing every figure to the digit at all five levels and confirming the asserted closed-form step, but the framing does not survive, since X is pinned by the identity X = M - Nbar + n_0 so the whole comparison collapses to one cell and the joint is measurably not a product elsewhere; the model error is 0.52 per cent, not 0.050. | [verify-cofactor-convolution.md](history/staging/verify-cofactor-convolution.md) |\n| X | `Q-xchan-at29-prereg` Does the joint-deficit closed form survive a blind test at @29? | OPEN | Pre-registration only, committed alone before any producer existed: the statistic, the predictions adopted verbatim from the record, two acceptance bands, the validation gate the instrument must clear before any @29 number is reported, and what each verdict does to TODO item X. | [xchan-at29-prereg.md](history/staging/xchan-at29-prereg.md) |\n| X | `Q-xchan-at37-offset` Does any registered offset-correction candidate for the ~3.8 law survive at @37? | ANSWERED | Sealed and committed alone before any @37 census of any kind existed, fixing the candidates, the sigma model and its projection band, the scoring rule and what each verdict does to item X's offset clause; scored in xchan-at37-score.md, where every registered candidate is killed and the number survives an independent recount. | [xchan-at37-offset-prereg.md](history/staging/xchan-at37-offset-prereg.md) |\n| X | `Q-xchan-at37-score` What does the @37 census say about the sealed X-channel offset pre-registration? | ANSWERED | The census measured 1 - J = 0.020823, below every registered prediction, so scored exactly as registered every one of the seven candidates dies at \\|z\\| = 104 to 122 and the survivor set is EMPTY, an outcome the sealed prereg has no consequence clause for; the number itself survives an independent recount on a different marking scheme, and the offset question is replaced by a larger one. | [xchan-at37-score.md](history/staging/xchan-at37-score.md) |\n| X | `Q-xchannel-at23` What is the fifth point of the X-channel statistic, at @23, and does the monotone rise hold? | PARTIAL | The fifth point is +0.2658 at @23, the rise holds at five levels and is larger than the four-point trend predicted, and the deficit has migrated into m >= 3, whose share of the X-gap runs 10.5, 23.5 and 81.1 per cent at @17, @19 and @23; the constant itself is not derived here. | [xchannel-at23.md](history/staging/xchannel-at23.md) |\n| X | `Q-xchannel-closedform` Does the zero-parameter closed form for the joint deficit survive its two blind tests at @29 and @31? | PARTIAL | 1 - J = 4S2 is the only one of the three pre-registered laws left standing and it is not exact: a clean hit at @29, z = -0.90, and 4.93 sigma low at @31 where sigma_J falls to 0.000024, with the two rivals dying at 6.35 and 20.81 sigma; both blind levels agree on a stable relative offset of about half a percent, -0.41% and -0.48%, whose absolute residuals match to 2%. The @31 figure is quoted at the run's own sigma_J = 0.000024, and audit #85 (accepted, verified, its also_fix unapplied) records that the note's section 7 calls that sigma inherited rather than derived: under the slot-clustered sigma calibrated in `item-x-offset.md` section 4 on seven known-truth controls the same residual reads z = -2.32, inside the pre-registration's own HIT band, so this row's \"one detection at 4.93 sigma\" is zero detections at the pre-registration's threshold; the adjacent `Q-xchannel-offset` row quotes -0.43 and -2.32 for this same law at the same two levels, a factor 2.09 and 2.13 apart. | [xchan-at29.md](history/staging/xchan-at29.md) |\n| X | `Q-xchannel-offset` What produces the ~3.8 law's -0.45% offset, and which candidate correction survives? | PARTIAL | The best available description is a finite-level correction of S3's size and sign - 1 - J = 4S2 - S3 scores z = +0.04 at @29 and -0.41 at @31 where the standing law scores -0.43 and -2.32 - but it is POST HOC; two candidate corrections and one shape are refuted on existing data, the sigma behind the @31 detection is calibrated on seven known-truth controls, and the deciding blind test at @37 has RUN: `Q-xchan-at37-score` (ANSWERED, 2026-08-21) measures 1 - J = 0.020823 there, below every registered prediction, so scored exactly as registered every one of the seven candidates dies at |z| = 104 to 122 with a 13.4*S3 shortfall and an EMPTY survivor set, and the finite-level-correction family is CLOSED as the offset's description, this row's 4S2 - S3 candidate with it; the @37 note's proposed TODO item X update is recorded there as \"report only, not applied\". | [item-x-offset.md](history/staging/item-x-offset.md) |\n| X | `Q-xchannel-triples` Does the joint deficit deepen or decay as the mixed super-W triple census is extended past @17? | ANSWERED | It decays: J = 0.9025, 0.9599, 0.9659 at @17, @19, @23, so the deficit runs 9.7%, 4.0%, 3.4%, both new points land inside the pre-registered DECAYING FASTER THAN FORCED band, and the DEEPENING instinct is refuted by 3.50 sigma between @17 and @19 and a further 1.68 sigma to @23. | [xchannel-triples.md](history/staging/xchannel-triples.md) |\n| A | `Q-anchored-ladder` Does the unified anchored ladder generalize past @11 and @13, and what does cheap K buy at each level? | ANSWERED | It generalizes verbatim to @17 and @19, and the K = 0 unified floor is DEAD there (vacuous at -1262 and -55865); what survives and grows is the family lead at the classic certification depth K*, margin +2, +5, +26, +110 across @11 to @19. | [attack-anchored-02.md](history/staging/attack-anchored-02.md) |\n| A | `Q-anchored-ladder-17-0830` At @17 and above, at which depth K does the composed (unified) anchored floor first turn positive and at which K does it reach the truth, and does the depth-cost curve K_first_positive(level), K_truth(level) move with level? | ANSWERED | At @17/@19 the answer was already on record (Q-anchored-ladder: positive at K = 2/10, truth at ascending K = 109/410, minimal pool 88/350) and is re-derived here as the custody gate; the new level @23 ran under a sealed 14-of-14 forecast: positive first at K = 27 = K* (PROVEN from the record before the run), truth at ascending K = 1732 of 1739, minimal pool 1543 = 0.887 of the scour (PROVEN lower bound meets VERIFIED upper bound); across five levels positivity costs exactly K* while the truth fraction rises 0.40, 0.62, 0.73, 0.81, 0.89, a MEASURED finite table with no law in x and no exponent touched. | [attack-0830-anchored-ladder-17.md](history/staging/attack-0830-anchored-ladder-17.md) |\n| A | `Q-anchored-unify` Can the three anchored mechanics be unified at @11, and what floor family does the unified cap certify? | ANSWERED | MEASURED and PROVEN-here but HELD: extending the staircase's cofactor injection to m = 1 folds the self-strike allowance into the count, the unified-cap lemma gives fresh <= capU_K <= cap_K at every depth and prime at @11 and @13, and the floor family runs 36 to 45; the x = 37 secondary is a clean negative. | [attack-anchored-01.md](history/staging/attack-anchored-01.md) |\n| A | `Q-history-dial` Does revealing the scour primes' classes concentrate the floor on a bounded set of strike locations, and what is the history dial worth? | ANSWERED | Disconfirming: the class axis is linear at 2.90 survivors per prime read, so no bounded set of strike locations carries the floor, and the touch-count ceiling is proven and tight; deliverable (1) of the brief already existed as canonical state (the forcing ladder 16 -> 45 at @11), so nothing new was bought there. | [attack-history-dial.md](history/staging/attack-history-dial.md) |\n| A | `Q-redteam-0830-engine` Do attack-0830-buchstab-deep.md, attack-0830-comb-tail.md and attack-0830-anchored-ladder-17.md survive an adversarial re-derivation on code that shares nothing with their producers, and does REFUTED row 97 stand as worded? | ANSWERED | Every headline number in the three notes reproduces exactly on independent code (26 of 26 comparisons, including F2/f2 = 3.6682 at sigma = 4.7088 by a near-analytic route, the 0-of-1512930 pair count, the tail defect at five levels, the full 419328-term Legendre split, and the whole @23 anchored ladder including K* = 27, floorU = 5364, truth 597475 and the minimal pool 1543 proven exact); one load-bearing DERIVATION is REFUTED — \"\\|B - 1\\| >= c forces sigma(q) < u_c - 1\" fails at @23 on 21134 (q, K) pairs for the engine's model B at c = 0.1 and on 33875 high-count pairs for the measured B at c = 0.01, because the step bounds every u in a weighted average by the largest one and treats the denominator as e^-gamma, so the \"level-free\" clause of buchstab-deep section 4 and REFUTED row 97's \"\\|B - 1\\| >= 0.01 forces sigma < 1.798\" must be re-graded from DERIVED to MEASURED-at-@23 (where the conclusion survives: the worst sigma is 2.159, still far below beta2 = 4.26645); row 97's \"the best sigma any level has\" is separately WEAKENED, since @29's head sigma is 5.5785 with width 1.5325 and @97's is 17.1422 with width 1.0000; comb-tail derives its main-term factor from prod (1 - 1/q_i) and measures it against prod (1 - 1/(q_i - 1)), which is what natal-cap-28 line 283 computes against its own line 20 and against certificate-engine.md section 1, a 0.9% difference at @23 that moves nothing certified; comb-tail also misquotes its own cited source as F1 in [1.5, 4] where that source says [1.5, 3] and says there is no bound at all at s < 1; four smaller slips are listed and none opens or closes a route. | [redteam-0830-engine.md](history/staging/redteam-0830-engine.md) |\n| 1 | `Q-at43-bigint-0830` Can the K-30 natal march be carried past its 2^53 ceiling to @43 exactly, and is the @43 point worth its cost? | PARTIAL | The engine side is done and VERIFIED (five paths promoted to BigInt, @7..@37 and a 0.248% slice of @41 reproduced digit for digit); the @43 point is NOT run, because the measured extrapolation is 579 h of eight cores on this machine (a 43x tile times a 2.23x per-cell cost, against the brief's ~13x), which is a weeks-class box job whose only payoff is a sixth point on a curve with no consumer; the forecast 0.8393 raw / 0.8399 persisted stands unscored. | [engine-0830-at43-bigint.md](history/staging/engine-0830-at43-bigint.md) |\n| 1 | `Q-redteam-0830-imports` Do the six lighter notes of 2026-08-30 (varE-identification, smooth-aps, rec-killrun, skeleton-door, at43-bigint, coherence) survive an adversarial pass on independent code and at the source pages, and which of their load-bearing claims are wrong? | ANSWERED | No note is refuted whole and no exponent moves. Every decisive measurement reproduces on code sharing nothing with the producers: the varE mixed remainder and its 0.844 / 0.826 share below 2L and 94.4 / 98.7 percent unbalanced share at x = 13, 17; the Buchstab deviations and all three crossing levels log10 z* = 33.53, 68.80, 33.62; the skeleton door's 9.2 / -0.9 percent, G30_agg 0.1113 / 0.1011 and the 0.0102 cap at @13, @17, by a path that also verifies cap-36 Theorem A; the @43 parity exactness on all 26,157,448 scour classes with 0 above 2^54 and a constructed @47 counterexample; and the five recon-0828 verdict counts summing to 75, which REFUTES rec-killrun's brief-error clause and confirms the orchestrator's rider. Seven claims WEAKEN. The load-bearing one is smooth-aps's \"the weight is unmet in print\": Harper's J. London Math. Soc. 112 (2025) e70293 proves a Barban-Davenport-Halberstam asymptotic for an ARBITRARY sequence on exactly that dyadic all-classes object, and his own footnote says the missing max over x' can be incorporated, so the NEAREST citation changes even though the step still does not close (the 2025 range is sqrt(2x) < Q <= x and its Theorem 2 route bars sieved sets by name). Also: Henriot Corollary 2 applies, but to the ERRATUM's statement and under two hypotheses the note leaves unstated (eps < alpha/600, not \"every eps\"; F at prime powers); varE section 4's display (*) carries an unfilled placeholder and is vacuous by a factor L/d; the skeleton door's \"would tighten\" is wrong in sign at @17, where the closable block is negative and removing it RAISES G30_agg to 0.1020; Halasz-Montgomery is not applicable for the reason given (it is an abstract inner-product inequality); FGKT's heuristic quote is on p. 4, not p. 3; and the \"~13x had no source\" is a corpus miss, the retired G2-walk figure at CHANGELOG.md:5716. Not TPC-strength; nothing here opens a route. | [redteam-0830-imports.md](history/staging/redteam-0830-imports.md) |\n| Z7 | `Q-zonegap-03-score` Do the ten predictions sealed in zonegap-03-prereg.md score against the stage-3 sweep at X = 1e12? | ANSWERED | All ten sealed rows score HIT and none miss, after a second pass added a band-edge argument to zonegap-01.js and re-ran the decade so the two rows that named an unprinted band could be scored; the four Group T hits are a custody promotion that follows from CUSTODY 1 and 3 passing, one sub-clause of T5 (the full-decade sd) stays unprinted, and the sweep still needs a DERIVED engine because zonegap-01.js's inlined 41-record ladder makes it exit at 1e12. | [zonegap-03-score.md](history/staging/zonegap-03-score.md) |\n| 2 | `Q-var41` What does the stable law predict for Var(41), and what can the tenth Var/E point pin? | OPEN | Pre-registration only, sealed and committed alone before any Var(41) engine exists: it freezes the prediction, a band taken from the law's own residuals at z <= 37, the derived z(41) prediction, and the honest statement that one more point cannot separate a limit from a drift. | [var41-prereg.md](history/staging/var41-prereg.md) |\n| 8 | `Q-applied-0828-engine` Were the engine red team's corrections applied to the five HELD notes? | ANSWERED | Applied, 37 edits across the five notes: two numbers corrected (first s >= 10.82 at x = 263 not 239, first non-empty kappa=1 level @37 not @53), two ledger verdict lines and one note title rewritten, one proof step given BV's max-over-y form, the one-class freshness factor restated on P^-(m) >= q so it holds at every scour prime, and two mislabelled columns renamed; six corrections that land on live documents are collected here unapplied, and no producer was touched. | [applied-0828-engine.md](history/staging/applied-0828-engine.md) |\n| 8 | `Q-applied-0828-live` Which engine and head red-team corrections reached the live layer? | ANSWERED | Nineteen edits across four documents: the glossary's cap_K dimension count with its mandatory sifted-versus-assumed clause plus three new entries (twin opener, head, tail); the survey's one-class density factor, S1's restored pi(q-1) terms and r in N_x hypothesis, the dropped Chen absorption, Theorem C with its emptiness, dimension 1 for q_i < q, the comb-conditioning qualifier and the inverted shallow-band bullet; the certificate engine's equidistribution status cell, the q = T^{o(1)} second hypothesis and the Comb Discrepancy scope cell; and the census's R as the continuum functional with its three comparators; paper/staircase-note.md, TODO.md and QUESTIONS.md are owed and untouched. | [applied-0828-live.md](history/staging/applied-0828-live.md) |\n| 8 | `Q-buchstab-deep-0830` Can the certificate engine's deep-ladder Buchstab transfer be proven at dimension 2 at the depths the run levels reach by a route that does not go through the fundamental lemma at s >= 22.06 or the constant-free DH Theorem 9.1 (TODO 8a)? | CLOSED | NO by any instrument this corpus can cite, and the non-closing step is structural, not a constant: the transfer is an asymptotic for a ratio of two dimension-2 sifting functions at one sifting depth z = q, so a sieve bracket [f2, F2] survives undivided in the ratio (width 3.67 at the best sigma any level has, 4.71 at the @23 head); a correction \\|B - 1\\| >= 1% can only occur at sigma(q) < 1.8 where f2 = 0 and F2 >= 7.85, while a lower bound exists only at sigma > 4.266 where \\|B - 1\\| < 1e-5; the Buchstab identity iterated once is already negative at K = 1 for q = 37 at @23 and is the sieve itself when iterated fully; Jurkat-Richert fails Omega(1) as stated; in the certificate's legal direction F2 beats the trivial cap_K <= cap2 at 0 of 1512930 (q, K) pairs at @23, so the sharp-sieve floor is -1733138 at every K; the transfer stays HEURISTIC, what would move it is a kappa = 2 asymptotic in the open band 2 < sigma < 4.266, and part (b) is scoped and graded, not attacked. | [attack-0830-buchstab-deep.md](history/staging/attack-0830-buchstab-deep.md) |\n| 8 | `Q-buchstab-transfer` Is the certificate engine's Buchstab transfer provable in the shallow regime y = T^{o(1)}? | ANSWERED | PROVEN conditional on the fundamental lemma and EMPTY at every run level (needs s >= 22.06; available 4.7 @23, 17.1 @97); the engine's B is a dimension-1 object, the survey's pair-Buchstab is dimension 2, the certificate needs a third. | [thm-buchstab-transfer-shallow.md](history/staging/thm-buchstab-transfer-shallow.md) |\n| 8 | `Q-comb-discrepancy` Can the Comb Discrepancy Lemma's constant 2*3^k be tightened (TODO 8c)? | PARTIAL | The lemma IS the trivial per-block bound; the attained optimum is max G - min G, a computed number per level (slack 29x @23, 54x @29, growing); term-by-term pricing lifts the certified head share to about 28-29 percent, not the 50 hoped for. | [comb-discrepancy-tight.md](history/staging/comb-discrepancy-tight.md) |\n| 8 | `Q-comb-tail-0830` Can TODO item 8's remaining part (b), prime-comb equidistribution in the tail regime, be proven at the depths the certificate needs, or at least measured and priced at the enumerable levels (8b)? | PARTIAL | Not proven and no exponent moves: the tail defect is measured exactly for the first time (signed -0.01% of the tail main term at @23, -0.00% at @29, unsigned 0.60% and 0.25%, smaller than the head's 1.00% in aggregate but 8 to 37% per prime at its worst), the Comb Discrepancy Lemma's method fails in the tail on its MAIN TERM (off by a derived factor e^gamma(1/u - q^(1-u)), measured aggregate 1.14x at @23) before its error term (10^27 x main), the prime-comb Legendre form has 2^(k+2) terms of which the short ones obey the lemma's <1 bound and the long ones are prime-in-AP discrepancies whose term-by-term sum is 11.2% of main at @23 against a signed -0.01%; the all-x form at K = 0 is Theorem C (PROVEN, limit only) and at the certificate's depth K* the non-closing step is a kappa = 1 asymptotic at sifting parameter s < 2, a proof gap on data that show B_tail = 0.998-1.000 at K* through @29; the full-depth half is 8(a), CLOSED. | [attack-0830-comb-tail.md](history/staging/attack-0830-comb-tail.md) |\n| 8 | `Q-redteam-0828-engine` Do the five 2026-08-28 certificate-engine notes survive an adversarial re-derivation, and does the one-class freshness factor correction hold before it reaches a live document? | ANSWERED | The one-class factor holds and is stronger than either note states; four numbers are refuted (first s>=10.82 is x=263 not 239, first non-empty kappa=1 level is @37 not @53, rho*g_max is off by one gap, the @17 certified share and per-term constant carry the wrong convention) and one proof step needs the max-over-y form of BV; the cluster's shared verdict stands. | [redteam-0828-engine.md](history/staging/redteam-0828-engine.md) |\n| 8 | `Q-redteam-0830-engine` Do attack-0830-buchstab-deep.md, attack-0830-comb-tail.md and attack-0830-anchored-ladder-17.md survive an adversarial re-derivation on code that shares nothing with their producers, and does REFUTED row 97 stand as worded? | ANSWERED | Every headline number in the three notes reproduces exactly on independent code (26 of 26 comparisons, including F2/f2 = 3.6682 at sigma = 4.7088 by a near-analytic route, the 0-of-1512930 pair count, the tail defect at five levels, the full 419328-term Legendre split, and the whole @23 anchored ladder including K* = 27, floorU = 5364, truth 597475 and the minimal pool 1543 proven exact); one load-bearing DERIVATION is REFUTED — \"\\|B - 1\\| >= c forces sigma(q) < u_c - 1\" fails at @23 on 21134 (q, K) pairs for the engine's model B at c = 0.1 and on 33875 high-count pairs for the measured B at c = 0.01, because the step bounds every u in a weighted average by the largest one and treats the denominator as e^-gamma, so the \"level-free\" clause of buchstab-deep section 4 and REFUTED row 97's \"\\|B - 1\\| >= 0.01 forces sigma < 1.798\" must be re-graded from DERIVED to MEASURED-at-@23 (where the conclusion survives: the worst sigma is 2.159, still far below beta2 = 4.26645); row 97's \"the best sigma any level has\" is separately WEAKENED, since @29's head sigma is 5.5785 with width 1.5325 and @97's is 17.1422 with width 1.0000; comb-tail derives its main-term factor from prod (1 - 1/q_i) and measures it against prod (1 - 1/(q_i - 1)), which is what natal-cap-28 line 283 computes against its own line 20 and against certificate-engine.md section 1, a 0.9% difference at @23 that moves nothing certified; comb-tail also misquotes its own cited source as F1 in [1.5, 4] where that source says [1.5, 3] and says there is no bound at all at s < 1; four smaller slips are listed and none opens or closes a route. | [redteam-0830-engine.md](history/staging/redteam-0830-engine.md) |\n| 8 | `Q-sharp-sieve-range` Do the sharp sieve functions (Jurkat-Richert, DHR) make the two empty certificate-engine theorems non-empty at run levels? | CLOSED | NO at finite level: DH Thm 9.1 carries no written constant and the crude fundamental lemma is strictly better; as limit statements kappa=1 is first non-empty @37 (a 19.8-wide bracket; @53 is the first narrow one, 1.656) and kappa=2 @23. | [thm-sharp-sieve-range.md](history/staging/thm-sharp-sieve-range.md) |\n| 8 | `Q-u2-engine-depth` Is the certificate engine's operative depth heading to u = 2, and does that make TODO item 8's payout form TPC-strength? | ANSWERED | It is not: the measured operative depth runs u = 4.191 at @17 down to 3.557 at @97 and is still falling, so the brief's conditional does not fire and item 8 is shown neither TPC-strength nor safe; a WALL-ADDRESS in the weakest sense, every figure SCRATCHPAD-GRADE. | [u2-engine-depth.md](history/staging/u2-engine-depth.md) |\n| Z0 | `Q-applied-0828-census` Were the census red team's corrections applied to the three HELD notes? | ANSWERED | Every WEAKENED and REFUTED verdict in redteam-0828-census.md is applied to its target note in the present tense; two producer defects and four live-doc corrections stay open and are listed here. | [applied-0828-census.md](history/staging/applied-0828-census.md) |\n| Z0 | `Q-applied-0828-quadpoint` Were the quadpoint red team's corrections applied to the four HELD notes? | ANSWERED | Applied: the nine REFUTED and WEAKENED verdicts that target the four notes are written into them in present tense, with ledger blocks added; the tenth targets the sealed prereg, outside the fence, and is recorded in attack-quadpoint-02.md instead; three corrections owed by TODO.md Z2 and one by Z0 are listed here and are NOT applied. | [applied-0828-quadpoint.md](history/staging/applied-0828-quadpoint.md) |\n| Z0 | `Q-redteam-0828-census` Do destroyer-census-01.md, stretch-01.md and records-placement-01.md survive an independent re-derivation of every load-bearing claim, and may they leave HELD? | ANSWERED | All three survive on arithmetic, every decisive number reproducing exactly on an independent engine; one sealed reading (records READ-3) is REFUTED as vacuous and four sentences need correcting before integration. | [redteam-0828-census.md](history/staging/redteam-0828-census.md) |\n| Z0 | `Q-redteam-0828-quadpoint` Do the four HELD quadpoint notes of 2026-08-22 survive an adversarial pass: the capture identity's proof line by line, the depth law and its asymptote, the sealed prereg scoring, and the prior-art citations at page level? | ANSWERED | The identity is PROVEN once its window convention is written down and holds at every K of 50 anchors under an independent implementation; four load-bearing sentences are wrong (section 4's gloss on Sum capU, v1 section 2's limit, the prior-art note's 2%, and the note's unconditional y* sentences), and none of the four defects opens or closes a route. | [redteam-0828-quadpoint.md](history/staging/redteam-0828-quadpoint.md) |\n| Z0 | `Q-review-request-0907` Which claims of the 2026-09-07 session should an independent math agent check, in what order, against which records, and what must the review return? | ANSWERED | Reviewed 2026-09-08 by an outside reader (history/reviews-0907/10): ranked claims 1 (conditional on the unread 1974 page), 2, 4, 5, 6 verified within stated scope; claim 3 verified with its quoted tolerances corrected (max deviation 1.8e-4, not 1e-4 or 1e-5); the Corollary 1 slip sharpened (fails for every representative other than 1, and mis-signs even then); ASSUMED items 4 and 6 discharged, 2 partially, 1, 3, 5 not. Twin-prime infinitude and every sufficient signed margin remain OPEN. Ranked below: the §6.2 demotion in the two-class manuscript (a change to a proof's dependency), the D_y identity reading (a change to what a census can show), the Erdős #687 attribution, and the two measurement artifacts. Twin-prime infinitude and every sufficient signed margin remain OPEN. | [review-request-0907.md](review-request-0907.md) |\n| A9 (retired) | `Q-a9-histogram-operator` What is the exact evolution of the gap-size histogram under folding, as a transfer operator on count(d), and what is the tail P(gap >= 2p)? | ANSWERED | The operator is derived and then verified by exact computation at every fold from T5 to T31 and the tail law is computed exactly, but the machinery is a rediscovery of Holt and Rudd arXiv:1408.6002 section 5 and may not be presented as new; only the G2 spacing beyond their \\|s\\| < 2p restriction is ours. | [a3-09-histogram-operator.md](a3-09-histogram-operator.md) |\n| G (retired) | `Q-block-L-first-dead` Is the block-1 counting bound on L the first-dead 62 or the last-alive 111, and does downward closure hold in the block coordinate? | ANSWERED | Downward closure HOLDS in the block coordinate: L <= 62 is sound and recomputed from scratch rather than inherited, L <= 111 is true, implied by 62 and weaker by 1.79x, and the adjudication's supposedly sharper instrument is attack 3's instrument. | [adjudicate-L-bound.md](history/staging/adjudicate-L-bound.md) |\n| 1c (retired) | `Q-c2prime-drift` Why does c2' drift on the 22-term ladder? | PARTIAL | Class-count-blind (attack-c2drift-01 §1); on item 1c's own band x >= 41 the ten-point rate is unresolvable: c2' 0.3745 +/- 0.1411 against c1's 0.3344 +/- 0.2358 on the same primes, a difference of 0.0401 = 0.12 of the control's ten-point window sd. Whether the drift is real is NOT closed here: attack-c2drift-01 measured it at p = 0.0053 on x >= 17 and left (i)' and (ii) alive. exponent-control §2/§4/§6 refit at 22 terms. | [attack-c2drift-01.md](history/staging/attack-c2drift-01.md), [c2prime-refit-22.md](history/staging/c2prime-refit-22.md) |\n| 11 (retired) | `Q-capK-bv` Is the prime-regime cap_K over the full wheel an unconditional asymptotic via the fundamental lemma over Bombieri-Vinogradov? | ANSWERED | PROVEN at short-note grade for q_K = W^{o(1)}, and EMPTY in every computable range (needs s >= 10.82, first cleared at x = 263; the lemma's floor s >= 10 is first reached at x = 239); the deep ladder is untouched. | [thm-capK-bv.md](history/staging/thm-capK-bv.md) |\n| 11b (retired) | `Q-consolidation-wave` Can the five cost-class scripts be bound to their own output, and can U-FRAME's two lost tables be rebuilt? | ANSWERED | All four runnable cost-class embeds are bound and the hand-pasted-tail counter falls 19 to 15 with the gate at TOTAL 0; three needed --force at a cost of 24 tokens out of 211 with no value changed anywhere, and natal-cap-33-overnight.js is a three-run composite that should never have been on the cost list. | [consolidation-wave.md](history/staging/consolidation-wave.md) |\n| 10 (retired) | `Q-excess-chain-c` Does the excess chain's constant c pin to 1.074 via a compressed-tail max correction? | CLOSED | NO: for any FIXED correlation length the derived family tends to 1 and both named corrections lower the maximum, so 1.074 is not derivable from a compressed-tail max correction; c measures 1.05 +/- 0.07 (sd) over x = 17..31, MEASURED, which does not separate 1.05 from 1.074. | [excess-chain-c.md](history/staging/excess-chain-c.md) |\n| 1b (retired) | `Q-greedy-oracle` Is the corrected greedy a G2 oracle? | ANSWERED | ORACLE ESTABLISHED, 13 of 13 exactly-known terms hit with minimum ratio 1.0000, so the greedy RULE is an exact solver where the answer is checkable at x <= 41; the rider matters more, the search budget needed grows 3.31x per additional prime and two objects whose truth reaches further show the same estimator's fidelity DECAYING, so nothing is established for the greedy AS RUN on the ladder. | [greedy-oracle-validation.md](history/staging/greedy-oracle-validation.md), [phase1-T1-greedy-oracle.md](history/staging/phase1-T1-greedy-oracle.md) |\n| 4 (retired) | `Q-import-fracparts` Does Saffari-Vaughan Theorem 10 cover the branches carrying the Skeleton door's mass? | ANSWERED | Prediction HIT: the covered branches carry at most 8 percent of the mass; anchor and a sharper wall-address banked; the THEOREM column is struck, not merely overpriced, because Theorem 10's range condition is x^{6/11+eps} < y, which fails at fixed M; no route; no cap-36 line is upgraded. | [import-fracparts.md](history/staging/import-fracparts.md) |\n| 0e (retired) | `Q-ioslack-survey` Does any instrument in the corpus carry a sharp-level signature that would let the infinitely-often slack be spent? | CLOSED | None does: 23 instruments surveyed and 19 eligible, smallest per-instrument p_FWE = 0.1082 and 0.8865 after Šidák, so dial 4's fourth channel closes and the dial is finished; the second finding is that maxsum_{L+1} is sharp exactly when L = 1. | [ioslack-survey.md](history/staging/ioslack-survey.md) |\n| Z5b (retired) | `Q-lambda-ledger` Does parity information survive in the lambda-weighted fold ledger? | CLOSED | REJECT, against a pre-registered threshold fixed before the producer existed: the lambda multiplier exists and is parity-free, so the ledger is blind at its design point rather than at its margins; two columns are pinned by proof, and nothing here is novel mathematics. | [attack-lambda-ledger.md](history/staging/attack-lambda-ledger.md) |\n| Z5b (retired) | `Q-ledger-forced` Which properties of the fold ledger's columns are FORCED rather than expected? | ANSWERED | Fourteen forced constraints, all identities or wrong-direction bounds; by_new is forced only in the prime gap (its O(1) is MEASURED); nothing forces net from below. | [fold-ledger-forced.md](history/staging/fold-ledger-forced.md) |\n| 11 (retired) | `Q-mod30-tail` Are the BV survey's S1 (mod-30 tail) and S2 (fixed truncation and depth) provable as stated? | ANSWERED | Both PROVEN at short-note grade (no second reader) once S2's freshness factor is corrected to one class per prime, which needs only P^-(m) >= q > q_i and so holds at every scour prime, not only in the prime regime; tail constant 2 ln 2 -> (ln 2)/2; nothing in the head. | [thm-mod30-tail.md](history/staging/thm-mod30-tail.md) |\n| 3 (retired) | `Q-monotonicity-sweep` Does anything in the corpus assume certificate validity is monotone in L? | ANSWERED | One unsound artifact and one invalid inference: attack-beta2-04-loss-budget.js section 6 bisects on a predicate measured not upward-closed and is wrong at 3 of 5 levels, true first-crossings 30/72/132/174/210 against the reported 36/72/144/174/354, and redteam-DP1-certificate.js draws a global minimality conclusion from a two-point local check; corrected, worst-casing certifies within 1.00 to 2.00 of true G2 and the exponent penalty runs 0 to 0.27 and falls with x. | [monotonicity-sweep.md](history/staging/monotonicity-sweep.md) |\n| Z6 (retired) | `Q-records-placement` Do the seven unswept twin-gap records 76-82, above 2^53, land uniformly inside their stretches, or is there square-anchor coupling in record-start placement? | ANSWERED | Sealed alone before the producer existed: the uniform band, the fixed definitions, the registered readings, the calibration abort gate and the riders are all fixed here and no placement fraction, q or stretch boundary was evaluated; scored in records-placement-01.md. | [records-placement-01.md](history/staging/records-placement-01.md), [records-placement-02.md](history/staging/records-placement-02.md), [records-placement-prereg.md](history/staging/records-placement-prereg.md) |\n| 5 (retired) | `Q-shadow-amplitude` Where does the kill shadow's drift amplitude come from, and does the record's counting floor explain it? | MIXED (shadow-amplitude-prereg.md: OPEN; shadow-amplitude.md: PARTIAL) | The finite-y correction is an identity with no free parameter and both measurement routes agree 10 of 10, but the record's own counting floor is four to six times too small and its candidate explanation is the wrong half, so the label stays PARTIALLY EXPLAINED. | [shadow-amplitude-prereg.md](history/staging/shadow-amplitude-prereg.md), [shadow-amplitude.md](history/staging/shadow-amplitude.md) |\n| 5 (retired) | `Q-shadow-prereg` Is the kill shadow's 0.85 the band-average of the Unification-Law survival curve over the post-crystallization window? | ANSWERED | Scored in shadow-buchstab.md: SHAPE-ONLY under the original y >= 997 rule; D3 passes in the reported ten clusters but D1 fails near y = 1000 at 1.49 times the 0.011 tolerance. Later finite-y normalisation and counting-floor corrections do not promote this verdict. The pair-Buchstab identification remains conjectural; the pre-registration body is preserved. | [shadow-prereg.md](history/staging/shadow-prereg.md) |\n| 0d (retired) | `Q-single-alignment` Does the head's single-alignment fold recursion for M(x, x^3) give a proven-shaped handle on the per-fold multiplier that the max-over-alignments recursion lacked? | CLOSED | The answer splits: the multiplier IS different in shape, measured - exactly 1 at 225 of 240 folds from x = 53, total spend 1.09 nats against a tile-shaped ladder's 7.12, measured M(1613) = 2220 against a tile-shaped 3.7e5 - but the route to a growth law through the recursion closes anyway, because the growth is initiated by the window's expansion into fresh ground, a boundary term that is the localized gap problem re-posed and that no fold reaches. | [localized-single-alignment.md](localized-single-alignment.md) |\n| 4 (retired) | `Q-skeleton-decide-0830` Should TODO item 4, \"Skeleton Equidistribution: restate or demote\", be restated around the modulus-W mass or demoted off the board, and what exactly is banked either way? | ANSWERED | DEMOTE. Over every scour prime the branches on which the door is a fixed-modulus question carry 9.2, -0.9, 0.2 and 0.5 percent of the skeleton at @13, @17, @19, @23 (the -10.8 and 5.5 percent on record were 9 and 2 percent subsamples), so the door as named removes at most 0.0102 from a G30_agg whose open part is 0.094 to 0.126; on the open side the phase never wraps and no equidistribution statement remains, only the inequality; nothing on the live board consumes G30_agg < 1/2; no first move exists that is not already ANSWERED. Banked regardless: the Skeleton Collapse Theorem (PROVEN, all x, all q) and G30_agg < 1/2 CERTIFIED at six levels @11..@29. | [decide-0830-skeleton-door.md](history/staging/decide-0830-skeleton-door.md) |\n| 4 (retired) | `Q-skeleton-door` Is the Skeleton Equidistribution Conjecture the blocker on the anchored calm? | ANSWERED | It is not the blocker as far as the door can be seen: Theorem A (Trapezoid Cancellation) is proven for all x, q and branches, @29 is certified as an exact integer inequality at G30_agg = 0.1176 with margin 0.3824 over 7,863 scour primes, and the decay-law shortcut is REFUTED, the six levels being flat within their own spread with every fit made strictly worse by adding @29. | [natal-cap-36-skeleton-door.md](natal-cap-36-skeleton-door.md) |\n| Z3 (retired) | `Q-stretch-structure` What is the stretch S_q = [q^2, q'^2) as an object, where does the Stretch Postulate sit against the Zone Postulate, and does the square anchor's QR structure buy any density advantage? | ANSWERED | The implication chain and the QR kill law re-derive independently arrow by arrow; the anchor's structure is real, deterministic and redistribution-only (exactly two killed offset classes at every prime anchor checked), so there is structure and no advantage, and SP sits strictly above the strong Zone Postulate. | [stretch-01.md](history/staging/stretch-01.md) |\n| 0e (retired) | `Q-verify-monotone-depth` Does the Monotone Depth obstruction, which retires dial 4, survive an adversarial re-derivation? | ANSWERED | NARROWED: every measurement in the record reproduces, several to the last digit, but the mechanism as named is false as a general law and the corpus already holds a counterexample to it; what survives is an accounting of three channels priced in tenths of a nat against a need of 6.700, 8.191 and 9.637 nats, so dial 4 becomes priced slack with one channel open and unpriced. | [verify-monotone-depth.md](history/staging/verify-monotone-depth.md) |\n| 1e (retired) | `Q-wrongdirection-audit` Are any of the ten live targets secretly TPC-strength, before the sessions are spent on them? | ANSWERED | The audit closes no target: two of the ten come out TPC-strength and three more carry a TPC-strength face under a quantifier they are likely to drift into, which is a labelling result and not a reason to drop them; item 1e is settled only because the answer was already on disk, and one route-blocking number in it is hand arithmetic and unstamped. | [attack-wrongdirection-audit.md](history/staging/attack-wrongdirection-audit.md) |\n| Z3 (retired) | `Q-z3-immune` Do the square anchor's immune offset classes (stretch-01 §4) carry a survivor surplus beyond the redistribution-only guardrail, do they attract the floor's first survivor, and does restricting the transplanted caps to them lower K*? | CLOSED | No, on all three. The preregistered surplus test is VOID on its own matched control (a 0.6% ratio-of-means bias present with and without the quadratic point); the preregistered placement test HOLDS at Z = +0.49; the design-free class test reads the guardrail's r/(r-2) over 3.4 million twin openers with a largest per-prime deviation of 0.11% and per-prime Z in -0.48..+1.66; and the restricted certificate, while lowering K* at 61% of anchors, has no certificate at all at 93 of 1219 and is beaten head to head by the QR-BLIND restriction that certifies 1219 of 1219. The refinement is closed. | [attack-z3-immune-01.md](history/staging/attack-z3-immune-01.md) |\n| Z (retired) | `Q-zonegap-reduction` What must be proven for Z2, what already is, and what does the proven head machinery give the zone? | PARTIAL | The Zone Restriction Lemma is proven and verified at five levels - the in-zone twin slots of T_p ARE the in-zone twin primes, so Z2(p) IS the whole-tile gap object restricted to the head window - the three-slack chain is graded honestly, the transfer inventory is listed item by item, and the deep end is proven an onset desert over 24 levels; section 5 lists what was not reached. | [zonegap-02-reduction.md](history/staging/zonegap-02-reduction.md) |\n\n## 2. Every question, by id\n\n| id | status | question | verdict | TODO | records |\n|---|---|---|---|---|---|\n| `Q-0c-holesweep` | ANSWERED | At a fold, how does maxsum degrade copy by copy as the two deleted residues sweep through their possible positions? | The full curve is on disk for nine folds (5->7 through 31->37, every copy) and it is a PALINDROME: the Mirror-Sweep Lemma is proven here and verified exhaustively, giving the fingerprint any kill-count-free bound must reproduce, but no such bound was produced by this pass. | 0c | [attack-0c-holesweep.md](history/staging/attack-0c-holesweep.md) |\n| `Q-0c0e-level-selection` | CLOSED | Can a residue-deleted maxsum bound that only has to hold at levels of the argument's choosing carry the weak Zone Postulate? | CLOSED, and the mechanism has a name: the level-selection licence is worth 0.3187 nats once and is falling, no selection rule beats its own null by anything worth having across T_11..T_29, and the BV/EH sweep supplies no moduli-selecting statement to spend. | 0c, 0e (retired) | [attack-0c0e-level-selection.md](history/staging/attack-0c0e-level-selection.md) |\n| `Q-DP1-mechanism` | ANSWERED | Does anything consume two-point data about a sifted set? | One consumer exists and is priced: the sharp degree-2 Boole-Frechet bound on empty windows is unconditional and concludes at every window position, but it over-certifies G2 with a log-log slope of +5.20 against x over x = 5..19, so the price of stopping at two-point data grows as a power of x rather than as a constant. | none | [attack-DP1-mechanism.md](history/staging/attack-DP1-mechanism.md) |\n| `Q-L-own-law` | ANSWERED | Does L, the adjacent-kill run length, have a law of its own level by level, and which abscissa is it linear in? | L has no law of its own: (L+1)*mbar(T_v) is approximately G2(y#) at every tile and top prime where both sides are exact, spread 1.0000 to 1.0838 over four tiles at y >= 23 with mean 1.0198, so the residue is quantisation, the abscissa question returns a calibrated negative, and the distance to 529 is priced rather than closed. | none | [attack-L-law.md](history/staging/attack-L-law.md) |\n| `Q-L-subadditivity` | ANSWERED | Is L sub-additive when two prime sets merge, as the intuition that L should be smaller for the combined set suggests? | Read literally the intuition is wrong by a one-line proof rather than an edge case: L is super-additive, exhaustively censused to nine primes; read as a statement about reusing primes it is right, and that reading is worth more than the refutation. | none | [attack-L-subadditivity.md](history/staging/attack-L-subadditivity.md) |\n| `Q-T2a-prediction` | PARTIAL | What are G2(43#) and G2(47#), predicted blind from the thirteen exact terms? | Registered blind: G2(43#) central 630, window 561 to 747, and G2(47#) central 720, window 653 to 868, with the method's calibration PASSING on the eight-term prefix; the 43# term has since been enumerated at 618, inside the window and below the central value, and 47# remains unmeasured, with the report's own largest risk being that c2' may be about to turn down. | none | [phase1-T2a-prediction.md](history/staging/phase1-T2a-prediction.md) |\n| `Q-W1a-applied` | ANSWERED | Were the unapplied findings for partition A, the paper suite and citation-heavy files, applied? | Thirty-three findings, 24 applied, 3 refused at a primary source (two of which would have written a falsehood into the corpus) and 6 handed back; the most consequential item was not on the list, since covering-dive's stated reason for dismissing Kalmynin-Konyagin was false by the corpus's own proof and the corrected reason had to be found rather than copied. | none | [phase1-W1a-applied.md](history/staging/phase1-W1a-applied.md) |\n| `Q-W3-literature` | ANSWERED | What do the four literature targets give: the Omega(kappa, L) quantifier, the Kalmynin-Konyagin transfer, Erdos-Ruzsa 1980, and the two-class upper bound? | The quantifier is over all pairs, confirmed verbatim in two independent sources, and every page and section citation in beta2-note checks out; the higher-consequence find is that covering-dive's stated reason for dismissing Kalmynin-Konyagin is false, since the proof only sees the size of Omega_p, and the two-class upper bound survives a second independent attempt as ABSENT. | none | [phase1-W3-literature.md](history/staging/phase1-W3-literature.md) |\n| `Q-X-upper-0829n` | PARTIAL | Can the rough-pair census X(K) on the stretch window be bounded above, unconditionally and with an explicit error term, by a dimension-2 upper-bound sieve, and how loose is the bound against exact X(K)? | Yes, as a theorem with nothing open (Selberg's Λ² inequality with a remainder bounded by the interval structure alone, VERIFIED at all 10908 anchor-depth pairs), but it is loose by 7.82 to 10.13 against exact X(K) at the band depths, decomposed as a sieve loss 1.78 to 2.53 times a structural loss 4.01 to 4.39 that no upper-bound sieve can remove, the window holds the level below z² (s_max 1.81 to 1.96) so the bound stops improving past p_K = 31 to 61, its main term is the sifting function's and not the census's, and the drift is a crossing quantity outside the legal half and is not evaluated. | Z2 | [attack-0829n-X-upper.md](history/staging/attack-0829n-X-upper.md) |\n| `Q-a144311-vocabulary` | ANSWERED | In the vocabulary the literature actually uses, is there a published upper bound for G2, and is beta2 still the best known kappa = 2 sifting limit? | The terms of art are the paired Jacobsthal function and Jacobsthal-type functions for polynomials, and in that vocabulary no published upper bound exists at any exponent, beta2 is unimproved since 2008, nothing is found on A144311's asymptotic growth, and an Erdős-Rankin lower bound is free; both load-bearing corpus claims survive. | none | [audit-a144311-vocabulary.md](history/staging/audit-a144311-vocabulary.md) |\n| `Q-a3-05-bound-L` | PARTIAL | Can L be bounded unconditionally from run consistency, and where does that stop? | Theorem B gives the first unconditional bound, L <= 1 + m*, sitting 0 to 3 above the truth at all eight measured folds, but it is of order G2(old)/p, which is linear in x on the fold diagonal; section 8 locates the wall at one large-deviation hypothesis H″ the residue side cannot reach, and Theorem C inherits it for kappa(m). | none | [a3-05-bound-L.md](a3-05-bound-L.md) |\n| `Q-a9-histogram-operator` | ANSWERED | What is the exact evolution of the gap-size histogram under folding, as a transfer operator on count(d), and what is the tail P(gap >= 2p)? | The operator is derived and then verified by exact computation at every fold from T5 to T31 and the tail law is computed exactly, but the machinery is a rediscovery of Holt and Rudd arXiv:1408.6002 section 5 and may not be presented as new; only the G2 spacing beyond their \\|s\\| < 2p restriction is ours. | A9 (retired) | [a3-09-histogram-operator.md](a3-09-histogram-operator.md) |\n| `Q-ab-coupling` | MIXED (attack-ab-coupling.md: PARTIAL; verify-ab-coupling.md: ANSWERED) | How much of the counting criterion's known slack does the rigidity of the two kill channels A = 0 mod p and B = -2 mod p close? | The pilot survives, every held number reproducing on an independent code path, and the one correction is in its favour: the depth-3 ceiling is 38 slots, not 39; the LP answers NO, its exact depth-2 optimum over the full polytope being 54 slots, identical to the pilot's, so it prices the idea and does not clear 51. | none | [attack-ab-coupling.md](history/staging/attack-ab-coupling.md), [verify-ab-coupling.md](history/staging/verify-ab-coupling.md) |\n| `Q-adversary-wave2` | ANSWERED | Do the four same-evening headlines (kill shadow, scanstat, sofic, Shearer) survive a refute-first adversarial pass? | Nothing BROKEN: sofic SURVIVES and the other three SURVIVE WITH CORRECTIONS, the corrections being one verdict label a grade too high by the record's own prereg rule, three false custody sentences and an oversold prediction, a misdescribed extremal witness, and an atomicity hypothesis that is load-bearing and not shown. | none | [adversary-wave2.md](history/staging/adversary-wave2.md) |\n| `Q-advmin-1113` | CLOSED | Can joint caps quantified over classes recover the staircase-to-truth gap at @11 and @13? | CLOSED at @11: the adversarial minimum for the whole family is 16, below the class-blind ceiling and 18 under the staircase floor of 34, which also refutes the task's own framing that it would sit between floor and truth; at @13 the same family is priced at advmin <= 152. | none | [attack-advmin-1113.md](history/staging/attack-advmin-1113.md) |\n| `Q-agent-start` | ANSWERED | What must an incoming agent read and preserve before continuing the reviewed arithmetic campaign? | Onboarding and document ownership only. The second dispatch is completed and independently reviewed on 2026-09-09. RESEARCH-EXECUTION.md owns current dispositions and candidate next obligations; RESEARCH-HANDOFF.md owns the updated exact contract. No assignment is started by that review. Twin-prime infinitude and required signed estimates remain OPEN. | C | [AGENT-START.md](AGENT-START.md) |\n| `Q-anchored-calm` | PARTIAL | What is proven about the anchored calm, and what is not? | The calm names a phenomenon and never a claim: the Mirror-Sibling, Fusion, Mirror-Phase Doubling and Minus-Half results are PROVEN at all x and all scour q, the phenomenon itself is MEASURED (the anchored rotation at rank 2 of 510,510 at @17), and two things stay unproven, of which the Skeleton Equidistribution Conjecture is one. | none | [anchored-calm.md](anchored-calm.md) |\n| `Q-anchored-ladder` | ANSWERED | Does the unified anchored ladder generalize past @11 and @13, and what does cheap K buy at each level? | It generalizes verbatim to @17 and @19, and the K = 0 unified floor is DEAD there (vacuous at -1262 and -55865); what survives and grows is the family lead at the classic certification depth K*, margin +2, +5, +26, +110 across @11 to @19. | A | [attack-anchored-02.md](history/staging/attack-anchored-02.md) |\n| `Q-anchored-ladder-17-0830` | ANSWERED | At @17 and above, at which depth K does the composed (unified) anchored floor first turn positive and at which K does it reach the truth, and does the depth-cost curve K_first_positive(level), K_truth(level) move with level? | At @17/@19 the answer was already on record (Q-anchored-ladder: positive at K = 2/10, truth at ascending K = 109/410, minimal pool 88/350) and is re-derived here as the custody gate; the new level @23 ran under a sealed 14-of-14 forecast: positive first at K = 27 = K* (PROVEN from the record before the run), truth at ascending K = 1732 of 1739, minimal pool 1543 = 0.887 of the scour (PROVEN lower bound meets VERIFIED upper bound); across five levels positivity costs exactly K* while the truth fraction rises 0.40, 0.62, 0.73, 0.81, 0.89, a MEASURED finite table with no law in x and no exponent touched. | A | [attack-0830-anchored-ladder-17.md](history/staging/attack-0830-anchored-ladder-17.md) |\n| `Q-anchored-unify` | ANSWERED | Can the three anchored mechanics be unified at @11, and what floor family does the unified cap certify? | MEASURED and PROVEN-here but HELD: extending the staircase's cofactor injection to m = 1 folds the self-strike allowance into the count, the unified-cap lemma gives fresh <= capU_K <= cap_K at every depth and prime at @11 and @13, and the floor family runs 36 to 45; the x = 37 secondary is a clean negative. | A | [attack-anchored-01.md](history/staging/attack-anchored-01.md) |\n| `Q-anchored-vs-global-gap` | CLOSED | Is the anchored gap A(v) = F(v) - v easier to bound than the global gap G2(v#)? | The separation is real and correctly measured but worth exactly one unit of exponent against a deficit of 2.2665, and the anchored quantity's only proven bound is beta_2 = 4.26645, the IDENTICAL theorem rather than a consequence, because the DHR sieve is position-uniform; so the anchored object sits further from its bound, and Proposition D is a closure at today's exponent only. | none | [attack-block-09-anchored.md](history/staging/attack-block-09-anchored.md) |\n| `Q-anchored-windows` | SUPERSEDED | How much does anchoring at the origin enrich a window against a random window, and for how long? | SPENT, live treatment is paper/anchored-note.md: Theorems A-C proven (A and B unconditional, C's constants HL-conditional), anchored enrichment bounded by e^{2gamma} = 3.172 and mortal, the head-window dip constant e^{2gamma}/4 = 0.79305 with 0.89 a finite-size blend; the kill shadow is no longer a separate object and section 3's peak sequence remains unreproduced. | none | [anchored-windows.md](anchored-windows.md) |\n| `Q-applied-0828-backfill-1` | ANSWERED | Which legacy notes in chunk 1 received ledger blocks? | All 86 notes in the chunk received a block and none was skipped; 85 distinct question ids were minted, no id was reused from QUESTIONS.md section 2, and two notes were chained onto one shared id. | none | [applied-0828-backfill-1.md](history/staging/applied-0828-backfill-1.md) |\n| `Q-applied-0828-backfill-2` | ANSWERED | Which legacy notes in chunk 2 received ledger blocks? | 85 of the 86 notes in chunk 2 carry a block; only GLOSSARY.md was left unindexed, as a vocabulary registry that poses no question, and one id (Q-records-placement) was reused from QUESTIONS.md section 2 with its question text copied exactly. | none | [applied-0828-backfill-2.md](history/staging/applied-0828-backfill-2.md) |\n| `Q-applied-0828-backfill-3` | ANSWERED | Which legacy notes in chunk 3 received ledger blocks? | All 86 notes in the chunk carry a block; none was skipped, one id (Q-quadpoint-transplant) is reused from QUESTIONS.md section 2 with its question text copied exactly, one id (Q-f-census) is shared by the two notes of the f-from-census chain, and seven blocks name a TODO item whose Ledger line does not yet list them. | none | [applied-0828-backfill-3.md](history/staging/applied-0828-backfill-3.md) |\n| `Q-applied-0828-backfill-4` | ANSWERED | Which legacy notes in chunk 4 received ledger blocks? | All 85 notes of chunk 4 carry a ledger block, none reuses an id already in QUESTIONS.md section 2, and none was left unclassified. | none | [applied-0828-backfill-4.md](history/staging/applied-0828-backfill-4.md) |\n| `Q-applied-0828-census` | ANSWERED | Were the census red team's corrections applied to the three HELD notes? | Every WEAKENED and REFUTED verdict in redteam-0828-census.md is applied to its target note in the present tense; two producer defects and four live-doc corrections stay open and are listed here. | Z0, Z4, Z6 (retired) | [applied-0828-census.md](history/staging/applied-0828-census.md) |\n| `Q-applied-0828-closures` | ANSWERED | Were the closures red team's corrections applied to the five HELD notes? | Applied. 47 prose corrections across the five notes; c2prime-refit-22 drops from CLOSED to PARTIAL and three ledger verdicts are rewritten; four producer READINGS blocks and eleven live-document lines carry the same defects and are listed here uncorrected, behind the fence. | 10 (retired), 1c (retired), 1d, Z5b (retired), 0 | [applied-0828-closures.md](history/staging/applied-0828-closures.md) |\n| `Q-applied-0828-engine` | ANSWERED | Were the engine red team's corrections applied to the five HELD notes? | Applied, 37 edits across the five notes: two numbers corrected (first s >= 10.82 at x = 263 not 239, first non-empty kappa=1 level @37 not @53), two ledger verdict lines and one note title rewritten, one proof step given BV's max-over-y form, the one-class freshness factor restated on P^-(m) >= q so it holds at every scour prime, and two mislabelled columns renamed; six corrections that land on live documents are collected here unapplied, and no producer was touched. | 11 (retired), 8 | [applied-0828-engine.md](history/staging/applied-0828-engine.md) |\n| `Q-applied-0828-glossary` | ANSWERED | Was the hyperuniformity conflation corrected? | In the vocabulary yes: GLOSSARY's single entry is split into Sub-Poisson window counts (Var/E), which keeps every measured value, and Hyperuniformity, which now carries Torquato's definition, the class I/II/III taxonomy, the tile's vacuous class I by periodicity and the sub-Poisson-not-hyperuniform reading at L = y^u; in the corpus no, one staging note is corrected here and ten further loose uses in nine .md files plus eighteen in seven scripts sit outside the fence with their replacement sentences listed below. | none | [applied-0828-glossary.md](history/staging/applied-0828-glossary.md) |\n| `Q-applied-0828-head` | ANSWERED | Were the head red team's corrections applied to the three HELD notes? | Applied. Every WEAKENED and REFUTED row of redteam-0828-head.md lands in the note it targets as 42 prose edits, and all three ledger verdict lines now carry the surviving chain; four corrections fall outside the fence and are owed elsewhere, the load-bearing one being destroyer-census-01 §6(b)'s R and its h/R = 1.09 -> 1.03. | Z4 | [applied-0828-head.md](history/staging/applied-0828-head.md) |\n| `Q-applied-0828-litimports` | ANSWERED | Were the lit/imports red team's corrections applied to the six HELD notes? | Applied, 64 substantive edits across the six notes: the Saffari-Vaughan range exponent is 6/11 in both import notes, the fixed-M THEOREM in import-fracparts is struck and the cap-36 MEASURED to PROVEN upgrade withdrawn, row 15 loses its theorem column and row 16 is regraded stale, and the six weakened sentences in the four literature notes now read as the audit requires; four claims are declined with reasons, and nine live documents plus two fenced staging notes are recorded here as owing corrections that this pass did not apply. | Z5, 4 (retired) | [applied-0828-litimports.md](history/staging/applied-0828-litimports.md) |\n| `Q-applied-0828-live` | ANSWERED | Which engine and head red-team corrections reached the live layer? | Nineteen edits across four documents: the glossary's cap_K dimension count with its mandatory sifted-versus-assumed clause plus three new entries (twin opener, head, tail); the survey's one-class density factor, S1's restored pi(q-1) terms and r in N_x hypothesis, the dropped Chen absorption, Theorem C with its emptiness, dimension 1 for q_i < q, the comb-conditioning qualifier and the inverted shallow-band bullet; the certificate engine's equidistribution status cell, the q = T^{o(1)} second hypothesis and the Comb Discrepancy scope cell; and the census's R as the continuum functional with its three comparators; paper/staircase-note.md, TODO.md and QUESTIONS.md are owed and untouched. | 11 (retired), 8, Z4 | [applied-0828-live.md](history/staging/applied-0828-live.md) |\n| `Q-applied-0828-quadpoint` | ANSWERED | Were the quadpoint red team's corrections applied to the four HELD notes? | Applied: the nine REFUTED and WEAKENED verdicts that target the four notes are written into them in present tense, with ledger blocks added; the tenth targets the sealed prereg, outside the fence, and is recorded in attack-quadpoint-02.md instead; three corrections owed by TODO.md Z2 and one by Z0 are listed here and are NOT applied. | Z0, Z2 | [applied-0828-quadpoint.md](history/staging/applied-0828-quadpoint.md) |\n| `Q-applied-0828-queue0827` | ANSWERED | Were the 2026-08-27 drafted corrections applied? | Eighteen edit entries applied across six files, the Wu band-label inference and the saturation claim being the load-bearing ones; six proposals declined, two on ownership, one because the brief forbids promoting its note, one as already carried in paraphrase, one on an unverified inferred theta-law, and one because two scratchpad notes disagree on the number. | none | [applied-0828-queue0827.md](history/staging/applied-0828-queue0827.md) |\n| `Q-applied-0828-registries` | ANSWERED | Which lit/imports corrections reached the registries? | 13 edits across six files, plus one appended changelog section: IMPORT-MAP gains rows 15 (no THEOREM, the column struck), 16 (STALE) and 17 (CLOSURE, two-sided) with its three counts recounted to seventeen rows; SEARCH-CONVENTIONS gains the Kourbatov-b, generalized-Dickman and VC-of-multiples rows with each channel's calibration state; import-map-construction's determinantal cell, natal-cap-36's separate-census sentence, one REFUTED row and one PRIOR-ART block are corrected; GLOSSARY, TODO, QUESTIONS, the five loose hyperuniformity documents and two staging notes are left for their holders and listed here. | Z5, 4 (retired) | [applied-0828-registries.md](history/staging/applied-0828-registries.md) |\n| `Q-applied-0828-registries2` | ANSWERED | Which queued registry corrections were applied? | 15 edits across four files: SEARCH-CONVENTIONS gains four §1 rows (the census as a pretentious correlation with the Chowla warning, Boolean analysis on the CRT product, covering a cyclic group by translates, the fractional parts of N/n) and five §3 rows, plus one pipe fix that was hiding a clearing phrase from the gate; import-rough-anatomy's λ₁ → λ₂ comparand is corrected in four places, 12–23× becoming 2.11 to 2.86×; lit-dickman-variance §6.3's \"row 15\" is disambiguated to the drafted row it means; IMPORT-MAP's circularity and payoff tallies are recounted and now sum to seventeen; REFUTED.md needed no edit and got none. | none | [applied-0828-registries2.md](history/staging/applied-0828-registries2.md) |\n| `Q-applied-0828-todo-prune` | ANSWERED | Which TODO items left the file under its charter after 2026-08-28? | Five left (11, 10, 1c, 1e, Z5b), nine were restated on what remains (Z2, Z4, Z5, Z6, Z7, Z0, A, 0, 1d, 4, 8), the rest stand; every retained Ledger line is verbatim, no claim's rung moved, and eleven notes now need `(retired)` on their todo lines before the gate is green. | none | [applied-0828-todo-prune.md](history/staging/applied-0828-todo-prune.md) |\n| `Q-applied-0828-varE` | ANSWERED | Were the Var/E red team's corrections applied to the four HELD notes? | Every WEAKENED and REFUTED verdict in redteam-0828-varE.md is applied to its target note in the present tense; four ledger verdicts are rewritten, the theta=1 row is raised to PROVEN, and seven live-doc sentences plus one producer reading stay open and are drafted here. | 9 | [applied-0828-varE.md](history/staging/applied-0828-varE.md) |\n| `Q-applied-0829-ledger` | ANSWERED | Were today's notes linked into the question ledger's TODO guard, the Z2 draft's last two non-fit lines fixed, and the covering-prune root corrected? | All three landed and none of them moves a result: nine of thirty notes read now name the TODO item their move executes and four TODO items carry the new ids, so `node research/qc.js` reports the ledger check clean at 0 findings; the Z2 draft's legal items 3 and 4 carry the audit's instrument rider and the quantifier-police clause the red team asked for; and the covering-prune producer now computes the root of a*ln(a/e) = 1 instead of quoting it, giving a = 3.5911214767 and a divergence constant of 7.1822 lnln x against the 7.19 the file carried, a 0.08 per cent change with nothing downstream consuming it. | none | [applied-0829-ledger.md](history/staging/applied-0829-ledger.md) |\n| `Q-applied-0829-measure-a` | ANSWERED | Were red team A's corrections applied to the notes it reviewed? | Of the nine APPLY items in redteam-0829-measure-a.md §3a-3c, seven were applied here as exact string edits, two (E1 and E5) were verified as already applied by the producing agent with a producer fix and a forced re-embed, one clause of E1's proposed wording is left unapplied and is recorded as owed, and the four live-document items of §3d plus the QUESTIONS.md regeneration are left for the orchestrator; node research/qc.js returns TOTAL 0 across all thirteen checks after the edits. | none | [applied-0829-measure-a.md](history/staging/applied-0829-measure-a.md) |\n| `Q-applied-0829-measure-b` | ANSWERED | Were red team B's corrections applied to the notes it reviewed? | 17 of 17 APPLY items whose target is a staging note are applied in place, 2 ledger verdict lines rewritten as the red team specified (measure-record-null2-0829 and zone-tail-02-0829), 3 HOLD/DECLINE items deliberately not applied, 5 live-document items left verbatim for the orchestrator, and 3 producer items recorded as OWED without touching any .js; the fast gate returns 0 findings after the pass, the single mid-pass finding having belonged to another agent's file, research/measure-g2z2-0829.js:432, and never to any edit made here. | none | [applied-0829-measure-b.md](history/staging/applied-0829-measure-b.md) |\n| `Q-applied-0829-measure-c` | ANSWERED | Were red team C's corrections applied to the notes it reviewed? | 22 of the 25 items in section 3 target the four HELD staging notes and all 22 are applied (21 as the exact old-text to new-text strings the red team wrote, 1 as a minimal carry because filling its column for all nineteen rows would be a fresh grading judgement rather than a citation), plus 1 directed edit to lit-scourfield-2008.md section 0 that red team section 1g names and section 3 gives no string for; 3 items are flagged HOLD and are left verbatim for the orchestrator (varE-theta2-proof.md, lit-smooth-divisors.md, SEARCH-CONVENTIONS.md); 1 producer item is OWED and untouched, since the producer's readings 13 and 14 still carry the M-P1 failure that E8 and E9 reverse in the note and correcting them needs a re-embed; node research/qc.js returned TOTAL 1 on the run straight after the edits, the one finding being research/measure-g2z2-0829.js:432 in another agent's file, and TOTAL 0 on the run after this note was written, with refs, quotes and ledger clean on both. | none | [applied-0829-measure-c.md](history/staging/applied-0829-measure-c.md) |\n| `Q-applied-0829-refuted` | ANSWERED | Were the four refuted-claims audits' APPLY row rewrites applied to REFUTED.md? | 25 of the 27 APPLY items are in the file, 1 more merged into a sibling's text at the shared boundary row, 1 deferred to the in-progress red team, 7 HOLD items untouched and listed here for Chris, 4 items owed outside REFUTED.md; the table still holds exactly 67 data rows and qc.js reports refs 0 and crosslinks 0. | none | [applied-0829-refuted.md](history/staging/applied-0829-refuted.md) |\n| `Q-applied-0829-registries` | ANSWERED | Were the registry debts of 2026-08-29 (IMPORT-MAP rows, the PRIOR-ART spectrum line, the escape labels, the REFUTED HOLD proposals) landed? | All four debts are discharged as far as a non-adjudicating pass may take them: IMPORT-MAP carries rows 20 and 21 at CLOSURE-plus-anchor and anchor-plus-wall-address with nothing opened, PRIOR-ART's Holt-Rudd attribution now names the printed spectrum and the SCRATCHPAD-GRADE 2.82/ln p mixing rate with priority stated as Holt and Rudd's, recon-0829-escapes.md section 6 carries all sixteen owed (i)/(ii)/(iii) labels of which nine are fresh judgements and three come out (ii), and three of the seven REFUTED HOLD proposals were applied as wording and calibration while four are left verbatim for Chris because they touch a producer, a canonical file, a cross-file number or the index's format; the table stands at 67 rows and qc.js is 0 on refs, crosslinks, calibration and search-convention. | none | [applied-0829-registries.md](history/staging/applied-0829-registries.md) |\n| `Q-applied-C` | ANSWERED | Which wave-2 consistency findings were applied to the paper suite? | Applied across the seven paper/ files with post-edit line numbers, plus a CHRIS-DECIDES list, handoffs to other partitions and one unresolved item: two scopes for the Cov_adj anticorrelation exception count, both now printed, one of which should be checked and the other dropped. | none | [applied-C.md](history/staging/applied-C.md) |\n| `Q-applied-G` | ANSWERED | Which wave-3 partition G findings were applied, and what did the pass find on its own? | Applied to the eleven owned files: the OEIS off-by-one confirmed and provable four ways, a factor of two found in the seam draft's Hardy-Littlewood heuristic, the L gap corrected in three files from a flat 0.58 ln p to the home's 0.58 to 0.95 ln p range, seven orphans swept, and one supplied answer refused as wrong before it reached a header banner. | none | [applied-G.md](history/staging/applied-G.md) |\n| `Q-applied-Q` | ANSWERED | Do the two OEIS drafts survive an instrument that recomputes them rather than reads them? | Two defects found in the G2 draft (the Maier-Pomerance conjecture cited in the wrong direction, and a no-upper-bound-published claim the repository itself contradicts) plus a third error class routed from partition P; every DATA term was recomputed and audit-numbers went from 37/39 to 76/78 checks. | none | [applied-Q.md](history/staging/applied-Q.md) |\n| `Q-applied-wave2-A` | ANSWERED | Which wave-2 partition A corrections were applied, and to which files? | Five jobs applied across G2-STATE.md, LOCALIZED-GAP.md, maier-matrix.md and origin-excess.md, with partition A's own qc columns at 0 on every check afterwards; handoffs to other partitions and one decision needing Chris are listed rather than guessed. | none | [applied-A.md](history/staging/applied-A.md) |\n| `Q-applied-wave2-B` | ANSWERED | Which wave-2 partition B corrections were applied to U-FRAME.md, ATTACKS3.md and gate-multiplies.md? | Applied with the U-FRAME section map preserved (128 inbound pointers, 47 on 5a, nothing renumbered), the strategic contradiction resolved, statistics corrected, one dead quotation removed and two further corrections to the record found on the way; section 12 lists what stayed unresolved. | none | [applied-B.md](history/staging/applied-B.md) |\n| `Q-applied-wave2-D` | ANSWERED | Were wave-2 partition D's findings applied to the eight entry-path documents? | Applied across all eight files with qc total 22 to 21 and one refs finding of its own (a framework bug), two transfers pairs word-diffed; the em-dash scope question and one GLOSSARY entry are flagged as judgement calls rather than omissions, and other partitions were live in the same tree so the deltas are not all this partition's. | none | [applied-D.md](history/staging/applied-D.md) |\n| `Q-applied-wave2-E` | ANSWERED | Which wave-2 partition E corrections were applied, and to which files? | Seven jobs applied across covering-dive, two-class-lower-bounds, PRIOR-ART, sift-limit-attack, theta-ladder, maxgap-law and dhr-verification, with refs and quotes at 0 in this partition before and after; one novelty verdict in a publication-track file is weakened by de Polignac 1849 and flagged for Chris. | none | [applied-E.md](history/staging/applied-E.md) |\n| `Q-applied-wave2-F` | ANSWERED | Which wave-2 partition F corrections were applied to the natal-cap family and its compound claims? | Applied across nine natal-cap files with two new homes created (anchored-calm.md and certificate-engine.md); the Fused-Window Calm Lemma was decomposed and its name retired, the X-limitation theorem's scope is confirmed a scope error, and the refutation is recorded as bigger than it had been reported. | none | [applied-F.md](history/staging/applied-F.md) |\n| `Q-applied-wave3-H` | ANSWERED | Were wave-3 partition H's U-FRAME sections 10-12 split into child documents? | Split and applied: U-FRAME falls 1,032 to 772 lines with three children created, but the corpus as a whole is 87 lines larger and that must not be reported as a shrink; refs and crosslinks stay at zero and no transfers finding exists between any stub and its child. | none | [applied-H.md](history/staging/applied-H.md) |\n| `Q-applied-wave3-I` | ANSWERED | Was the flagship paper split as wave 3 partition I directed, and what did the split expose? | Split into paper/moire-primes.md and the new paper/wall-note.md, with every deliberate change to moved text word-diffed; three open defects sit inside the split, two of them already landed and one not, and the judgement calls are listed so they can be overruled. | none | [applied-I.md](history/staging/applied-I.md) |\n| `Q-applied-wave4-P` | ANSWERED | Which calibration defects sit on a3-05-bound-L.md, localized-04-maxsum.md and level-ledger-tight.md, and were they fixed? | Fourteen defects found and fixed on the one axis wave 3 never ran, none of them reachable by any check in qc.js: two numbers contradicting the file's own pasted script output, one slot count against thirteen other sites and an exact closed form, one superseded sifting limit and three dropped scopes; the gate stayed at zero on all six checks. | none | [applied-P.md](history/staging/applied-P.md) |\n| `Q-at43-bigint-0830` | PARTIAL | Can the K-30 natal march be carried past its 2^53 ceiling to @43 exactly, and is the @43 point worth its cost? | The engine side is done and VERIFIED (five paths promoted to BigInt, @7..@37 and a 0.248% slice of @41 reproduced digit for digit); the @43 point is NOT run, because the measured extrapolation is 579 h of eight cores on this machine (a 43x tile times a 2.23x per-cell cost, against the brief's ~13x), which is a weeks-class box job whose only payoff is a sixth point on a curve with no consumer; the forecast 0.8393 raw / 0.8399 persisted stands unscored. | 1 | [engine-0830-at43-bigint.md](history/staging/engine-0830-at43-bigint.md) |\n| `Q-attacks-wall-ten` | ANSWERED | Which ten attacks on the frontier-zone wall were executed, and what did each return? | All ten are marked DONE with their learnings recorded; none breached the wall, parity stands, and the durable yields are attack 8's unconditional small-gap theorem for p >= 17 and the anchored-origin phase picture. | none | [ATTACKS.md](ATTACKS.md) |\n| `Q-attacks2` | ANSWERED | What did the ten copy-geography attacks of the second wave find? | All ten complete: one theorem-grade result (the Exact Invariance Lemma, fossil depth frozen at birth, proved independently twice) and one MEASURED, Hardy-Littlewood-conditional law (the Unification Law, to about 1 percent); no unexplained novelty survives and the wall stands. | none | [ATTACKS2.md](ATTACKS2.md) |\n| `Q-attacks3-uframe` | ANSWERED | What did the ten-attack u-frame wave (A1-A10) settle about the fold structure and the Zone Postulate? | The R/I split held: every residue attack returned an exact object and no interval attack moved; the additive chain G2(new) <= G2(old) + L*mbar it was framed around is itself REFUTED (U-FRAME 5a step 4), A5's Theorem B is structurally capped, and A10's exactness route is a dead end (rate 0.927 over 329 cells, excess 6..120 not shrinking). | none | [ATTACKS3.md](ATTACKS3.md) |\n| `Q-audit-campaign` | ANSWERED | Does the campaign archive layer - the natal-cap notes, ATTACKS, covering-dive and their neighbours - hold internal inconsistencies? | Six files and eight of the natal-cap notes are clean and need no change; the remainder carry per-file findings recorded in place, and this is a diagnosis pass rather than an edit. | none | [audit-campaign.md](history/staging/audit-campaign.md) |\n| `Q-audit-cross-doc` | ANSWERED | Where does the corpus carry two values for the same object in two different documents? | Twenty-six findings ranked A, B and C, eighteen of them the Kourbatov shape where the resolution was already on disk in a document the stale one cites by name, including a retracted theta column quoted as live in nine places and a prior-art collision on the dimension-2 Rankin accounting; the gate read 0 findings before and after. | none | [audit-cross-document-constants.md](history/staging/audit-cross-document-constants.md) |\n| `Q-audit-front-door` | ANSWERED | Does someone arriving after the 2026-08-19/20 wave get the true current picture from README, G2-STATE section 0 and CHRONICLE in ten minutes? | All three are STALE: README's Status is dated 2026-08-17 with a broken map row and four missing files, G2-STATE section 0 stops at the second attack wave, and CHRONICLE ends at Day 5; drafts are ready for two of the three and README's rewrite is left in Chris's register. | none | [audit-front-door.md](history/staging/audit-front-door.md) |\n| `Q-audit-new1` | ANSWERED | Do G2-STATE.md, LOCALIZED-GAP.md, localized-04-maxsum.md, exponent-control.md and gate-multiplies.md agree with themselves and with their sources? | Findings per file, the leading one a NARROWED Origin Excess claim: the lemma carries an unstated third hypothesis, positivity of D_y(y'^2) - 2(pi(x) - pi(y) + 1), which forces ln x*/ln y near 1.44 at every level computed and caps the advantage at about 2.2 rather than order (ln x / ln y)^2. | none | [audit-new1.md](history/staging/audit-new1.md) |\n| `Q-audit-new2` | ANSWERED | Are the seven research documents of the new2 partition internally consistent and correctly scoped? | Per-file findings across the seven, mostly dropped scopes and attribution defects, plus four documents that were missing the standard current-understanding footer, which was added. | none | [audit-new2.md](history/staging/audit-new2.md) |\n| `Q-audit-numbers` | ANSWERED | Does the corpus's arithmetic survive recomputation from scratch by an independent method? | The ladder is right: all twelve G2 terms reproduce under a method sharing no code with the one that produced them, 37 of 39 assertions pass in 152 s, and the four errors found are small and none is load-bearing. | none | [audit-numbers.md](history/staging/audit-numbers.md) |\n| `Q-audit-papers` | ANSWERED | Are the paper suite and the root README consistent with their research homes? | Per-file corrections applied across the six paper/ files and README, with a Not fixed section naming what was deliberately left, including the exposition question the Holt attribution raises and one stylistic sentence of Chris's that was left alone. | none | [audit-papers.md](history/staging/audit-papers.md) |\n| `Q-audit-reading-path` | ANSWERED | Can a cold reader reach every live result in two hops? | The two-hop test passes for eight of the ten headline results of 2026-08-19/20 and no staging record is orphaned; what fails is narrower, the front door being three days behind and contradicting REFUTED.md on u_sup, G2-STATE's section 10 table still the 2026-08-17 list, and kappa-not-L carrying nothing from the L wave. | none | [audit-reading-path.md](history/staging/audit-reading-path.md) |\n| `Q-audit-spine` | ANSWERED | Do the spine documents - ZONE-POSTULATE, THE-DIALS, THE-LENS, GLOSSARY, OBSERVATIONS - state their objects correctly? | Findings per file, the leading one a frame error in ZONE-POSTULATE section 5: G2's own exponent measures 1.57 central, bracket 1.3 to 1.9, against a calibrated one-class control, while theta is above 2 and rising, so the two readings that sat on boundary exponent 2 resolve in opposite directions. | none | [audit-spine.md](history/staging/audit-spine.md) |\n| `Q-audit-staleness` | ANSWERED | How much of the live layer is stale against changelog passes 26 to 48? | Eleven of the nineteen files in scope are clean and eight carry at least one stale sentence (U-FRAME three, G2-STATE two and others), with a separate list of upgrades the live layer has not absorbed; report only, nothing outside the report was edited. | none | [audit-staleness.md](history/staging/audit-staleness.md) |\n| `Q-audit-uframe` | ANSWERED | Do the u-frame documents' claims survive an audit against their own sources? | Several are REFUTED: the multiplier requirement ln c <~ 2 ln^2 u / u is off by a factor ln u and is replaced by the per-fold rate ln c(p) <= 2 ln p / p, the Inertness Lemma's rate is narrowed to 4/u, and SUM L*mbar ~ 2 x ln^3 x is false because U-FRAME step 4 is false. | none | [audit-uframe.md](history/staging/audit-uframe.md) |\n| `Q-b2mean` | ANSWERED | Can the mean value of B2 = Sum e^2 Vabs(e)^2 be bounded below its trivial qmax*B ceiling? | Yes, by one logarithm: B2 = O(qmax*ln^7 z) against the trivial O(qmax*ln^8 z), with the mass located on primorial-core smooth e nowhere near qmax, the primitive-mode weight J2(e)/12 proved and priced as a flat 1.48x cut of MS3, and the finite forms computed exactly at z = 13..47. | none | [attack-b2mean-01.md](history/staging/attack-b2mean-01.md) |\n| `Q-barrier-kappa2` | CLOSED | Does Granville's exceptional-character construction, which makes the Jurkat-Richert f and F attained by interval sieving problems at kappa = 1, transfer to kappa = 2 and supply the missing barrier on wall-note Face 4, so that no class-blind argument can beat beta_2 = 4.26645 on intervals? | No. Verdict (b): the transfer runs, fixes the axiom obstruction that sift-limit-attack.md section 2 blamed for the mixed-sign sketch's death, and then hits a hard ceiling at u = 2, because the character's only gift is the one-point identity lambda = chi on z-rough integers and a sign pattern with a +1 coordinate is empty exactly when z^2 >= x; above u = 2 the surviving question is the shifted intersection of two positive-density lambda-classes, which is Chowla-strength, so the section 2 verdict stands with a corrected reason; the band (2, 4.26645] is untouched, and the hypothesis itself implies the target by Heath-Brown 1983. | 0 | [attack-barrier-kappa2.md](history/staging/attack-barrier-kappa2.md) |\n| `Q-bc-parity-floor` | CLOSED | Is the 2 that the census's B/C ratio limits to the same object as the parity floor's 2? | REFUTED at the first step, before any kappa-scan: the certificate dies where B/C crosses 1, not 2 (B = 503 against C = 497 at the death zone), the constant reads the sifting dimension while the threshold does not, and the parity floor's 2 cannot be this one. | none | [attack-bc-parity-floor.md](history/staging/attack-bc-parity-floor.md) |\n| `Q-beta2-AB-bounded` | ANSWERED | Is B(z,s) bounded, and does bounding it move the 4.2665 exponent? | B(z,s) <= 9 A(z)^2 (E(z) - 1) = O((log z)^8) is proven, unconditional, explicit and uniform in s, but the gap it closed was never load-bearing, the estimate has a published neighbour at s >= 9, Opera de Cribro Lemma 6.18, which does not cover this corpus's s in [2.0, 3.4], and the route to B = O(1) is measured dead as a uniform statement. | none | [attack-AB-bounded.md](history/staging/attack-AB-bounded.md) |\n| `Q-beta2-exact-strata` | CLOSED | Does re-inserting exact Buchstab strata (DP3) lower the dimension-2 positivity threshold below beta2 = 4.26645? | A sharp negative with a price tag: exact strata are legitimate and worth 1.72 to 1.84 of exponent, taking the positivity threshold to 2.0260 at finite levels, but they produce no theorem because the exact stratum value is the upper sieve function of a dilated tile one level down, so the regress terminates at every finite z and at no uniform z. | none | [attack-beta2-03-exact-strata.md](history/staging/attack-beta2-03-exact-strata.md) |\n| `Q-beta2-loss-budget` | ANSWERED | Where in the chain from truth to beta2 = 4.26645 is the loss, and which discard point is worth attacking? | Aim at theta: it carries the largest derivative (-4.2665 per unit) and the only consumer in the literature, beta is worth at most 1.07 and has no consumer, DP2 and the o(1) are worth exactly zero, and the two live knobs together stop at 1.597, so the last factor of 1.6 down to truth is in neither of them. | none | [attack-beta2-04-loss-budget.md](history/staging/attack-beta2-04-loss-budget.md) |\n| `Q-beta2-theta-total` | CLOSED | What is the largest theta_total defensible position-uniformly, and does it clear the break-even the beta2 upper-bound attack needs? | The largest defensible theta_total is 1 and it does not clear break-even: it misses by 0.2417 under the brief's symmetric arithmetic and by 0.2090 under the corrected asymmetric one, no part of the gap closes, and the weakest link in the negative is an ABSENT rather than a PROVEN. | none | [attack-beta2-02-theta-total-minimal.md](history/staging/attack-beta2-02-theta-total-minimal.md) |\n| `Q-beyond-chebyshev` | PARTIAL | Can the ensemble bound on P(S = 0) be pushed beyond Chebyshev, and how far? | Two beyond-Chebyshev bounds now stand, at @11 where capacity gives P(S = 0) = 0 exactly and at @13 where P(S = 0) <= 1.898e-6, a factor 513 below Chebyshev's 9.74e-4, with 683 assertions machine-checked; @17 is blocked on a 4.9e16-term sum and needs a 4-point wrap identity that does not exist. | none | [natal-cap-21-beyond-chebyshev.md](natal-cap-21-beyond-chebyshev.md) |\n| `Q-bf-split` | ANSWERED | Is Brüdern-Fouvry's level split optimal for their own theorem, or is there unclaimed exponent in it? | It was 99.3 per cent a trap: their condition (iii) binds at the published split and the proposed re-split violates it by x^{0.040403}, leaving 0.0000120536 in theta of the claimed 0.0295482779, real but invisible at the four decimals they print. | none | [attack-bf-split.md](history/staging/attack-bf-split.md) |\n| `Q-bilinear-fold-first-attack` | ANSWERED | Does assigning each Mobius factor to its first fold, or applying a direct second-moment bound, establish the missing one-sided estimate? | The first-fold identity yields no saving; a direct second moment has a negligible diagonal and an open two-prime off-diagonal. Fixed-fold covariance is an exact residue-imbalance square and the beta weights balance at each fixed modulus. Growing-depth joint control remains OPEN; B>=-C2*x+o(x) is not improved. | C | [bilinear-fold-attack.md](bilinear-fold-attack.md) |\n| `Q-bilinear-transplant` | CLOSED | Does this programme's frame have anywhere to put bilinear or Type-II arithmetic information, the only thing that defeats parity in practice? | The tile side cannot host it, since no corpus object is a signed sum over two ranging element ranges; the zone side can express one, and the estimate that one needs is the one already recorded absent. Section 9 lists the four findings that would reopen the row. | none | [attack-bilinear-transplant.md](history/staging/attack-bilinear-transplant.md) |\n| `Q-block-L-first-dead` | ANSWERED | Is the block-1 counting bound on L the first-dead 62 or the last-alive 111, and does downward closure hold in the block coordinate? | Downward closure HOLDS in the block coordinate: L <= 62 is sound and recomputed from scratch rather than inherited, L <= 111 is true, implied by 62 and weaker by 1.79x, and the adjudication's supposedly sharper instrument is attack 3's instrument. | G (retired) | [adjudicate-L-bound.md](history/staging/adjudicate-L-bound.md) |\n| `Q-block-alternation` | CLOSED | Is there a block-level alternation law? | No: a block-level inequality exists and is a real theorem, but it is not an alternation law - over a block the alternation law is strictly weaker than plain slot counting by the exact factor mbar(T_v)/6, at least 10/6 and growing like ln^2 v, and the resulting Block Partition Inequality closes only at v = 5 and v = 7 and is vacuous from v = 11 because 2*Sum_{v v^2 block ladder to sustain itself? | No: on the eleven-term exact ladder (this attack computed G2(41#) = 546) the ratio is flat at slope +0.05 +/- 0.11 on b >= 11 against a 12.3 per cent band, while the same estimator resolves the one-class control's falling ratio at 11.6 sigma; and the criterion is not independent, since at the natal it IS G2(x#)/x^2, the Zone Postulate itself. | none | [attack-block-01-ladder.md](history/staging/attack-block-01-ladder.md) |\n| `Q-block-exponent` | CLOSED | Does the block sub-problem have a better exponent than the proven 4.2665? | The required exponent for the ladder to close is 2, not 4 (the brief's step-3 arithmetic carried a factor error), and 2 is the TPC-equivalent line, so the useful part of the open band (2, 4.2665] is empty; the block's own prime range pins kappa = 2 and starting from T_v does not change the dimension. | none | [attack-block-05-exponent.md](history/staging/attack-block-05-exponent.md) |\n| `Q-block-greedy` | ANSWERED | Does the block-restricted adversarial greedy reach the truth at v = 5, and does block restriction cost anything? | Deterministic greedy reaches 17 against truth 19, randomised restarts reach 19 and an exhaustive DP confirms it, so greedy as a rule is optimal at v = 5 and the shortfall is a search failure; block restriction costs nothing on ten points with the same estimator on both sides, and the published Y2 ladder's bisection on a non-monotone predicate is a real defect. | none | [attack-block-04-greedy.md](history/staging/attack-block-04-greedy.md) |\n| `Q-block-kappa` | CLOSED | What does the block version of kappa(m) buy over the per-fold form? | Nothing: kappa_block(1) = 19 equals combined L at block 1 by construction, the minimal shift j*(m) equals kappa_block(m) at every m from 1 to 12 so there is zero slack in the block object, and the block form of Theorem C is empty, its vacuity coefficient 1.552 against a requirement of 1 at block 1 and growing like ln^2 v. | none | [attack-block-07-kappa.md](history/staging/attack-block-07-kappa.md) |\n| `Q-block-killers` | ANSWERED | For a maximal run of deleted T_v slots under the block v -> v^2, which primes do the deleting and how concentrated is the work? | Not concentrated: across 40 runs in five datasets every block prime kills at least one slot of every run and the exact minimum set cover is 82 to 100 per cent of the block, so no bound ranging over the block gets a reduction; the one favourable number is 915 of 1146 slots, 79.8 per cent, with exactly one killer. | none | [attack-block-02-killers.md](history/staging/attack-block-02-killers.md) |\n| `Q-block-literature` | ANSWERED | Does the literature already own the range-restricted covering object behind the block v -> v^2? | FOUND, and it has a name: the range-restricted covering problem is Erdős Problem #688 and is OPEN, the bounded-capacity conjecture #1200 is a third trap of the same class as the brief's two, and the Kalmynin-Konyagin journal version that covering-dive.md says does not exist does; the remaining negatives are ABSENT-AFTER-COMPETENT-SEARCH on calibrated channels. | none | [attack-block-06-literature.md](history/staging/attack-block-06-literature.md) |\n| `Q-block-reformulation` | CLOSED | Does the v to v^2 block reformulation, treating the block as a set, give a smaller combined L and an exponentially growing window? | Sound reformulation, no open path: no primality input is needed anywhere, nothing is deleted inside the window at all so combined L there is identically zero, and the object has to live in the tile; this note's own correction (the counting bound is 111, not 62) was itself reversed the same day by adjudicate-L-bound.md. | none | [attack-block-00-ADJUDICATION.md](history/staging/attack-block-00-ADJUDICATION.md) |\n| `Q-block-secondmoment` | CLOSED | Can a second moment over a block bound the empty windows, and is it allowed to work? | NO, on a tile-independent impossibility: Chebyshev needs dist(E, Z) < E/sqrt(W) and the measured shortfall at H = 204 is a factor 8.41e6, so a second moment gives a density bound that is legal at every H and TPC-relevant nowhere; the TPC line sits at eps = 1/W and at H about w^2, and quantifying over w buys nothing. | none | [attack-block-08-secondmoment.md](history/staging/attack-block-08-secondmoment.md) |\n| `Q-block-target` | CLOSED | What exponent does the squaring ladder v -> v^2 actually need, and do the steps get easier up the ladder? | The required exponent is exactly 2 with a constant below 1, which by the p^2 rule is TPC-equivalent, and the requirement is exponent-invariant at every step, so as a route to a theorem weaker than TPC the ladder is dead. | none | [attack-block-10-target.md](history/staging/attack-block-10-target.md) |\n| `Q-bonferroni-degree` | CLOSED | What degree does the Boole-Frechet certificate need, and does raising the degree reach an asymptotic bound? | Dead for asymptotics on a loop that closes on itself: raising the degree divides the over-certification exponent by the number of pairs without removing it (measured slopes 0.5441, 0.5425, 0.2816, 0.2937, 0.1695 at k = 2..6 against the predicted 1/(2 floor(k/2))), and Theorem A gives k >= (theta(x) - O(1))/(log G2 + O(1)), so the better the G2 bound assumed the higher the degree needed. | none | [attack-bonferroni-degree.md](history/staging/attack-bonferroni-degree.md) |\n| `Q-buchstab-deep-0830` | CLOSED | Can the certificate engine's deep-ladder Buchstab transfer be proven at dimension 2 at the depths the run levels reach by a route that does not go through the fundamental lemma at s >= 22.06 or the constant-free DH Theorem 9.1 (TODO 8a)? | NO by any instrument this corpus can cite, and the non-closing step is structural, not a constant: the transfer is an asymptotic for a ratio of two dimension-2 sifting functions at one sifting depth z = q, so a sieve bracket [f2, F2] survives undivided in the ratio (width 3.67 at the best sigma any level has, 4.71 at the @23 head); a correction \\|B - 1\\| >= 1% can only occur at sigma(q) < 1.8 where f2 = 0 and F2 >= 7.85, while a lower bound exists only at sigma > 4.266 where \\|B - 1\\| < 1e-5; the Buchstab identity iterated once is already negative at K = 1 for q = 37 at @23 and is the sieve itself when iterated fully; Jurkat-Richert fails Omega(1) as stated; in the certificate's legal direction F2 beats the trivial cap_K <= cap2 at 0 of 1512930 (q, K) pairs at @23, so the sharp-sieve floor is -1733138 at every K; the transfer stays HEURISTIC, what would move it is a kappa = 2 asymptotic in the open band 2 < sigma < 4.266, and part (b) is scoped and graded, not attacked. | 8 | [attack-0830-buchstab-deep.md](history/staging/attack-0830-buchstab-deep.md) |\n| `Q-buchstab-transfer` | ANSWERED | Is the certificate engine's Buchstab transfer provable in the shallow regime y = T^{o(1)}? | PROVEN conditional on the fundamental lemma and EMPTY at every run level (needs s >= 22.06; available 4.7 @23, 17.1 @97); the engine's B is a dimension-1 object, the survey's pair-Buchstab is dimension 2, the certificate needs a third. | 8 | [thm-buchstab-transfer-shallow.md](history/staging/thm-buchstab-transfer-shallow.md) |\n| `Q-bv-import-survey` | ANSWERED | Which unproven legs of the tile programme become theorems if Bombieri-Vinogradov, Elliott-Halberstam or GRH is imported? | Two legs become provable now, the fixed-modulus tail refinements S1/S2 by Siegel-Walfisz and the full-wheel prime-comb equidistribution in the tail regime by BV; the Buchstab transfer at bounded u and Assumption A for d = 2 are parity-blocked at every one of the four inputs. | none | [bv-import-survey.md](bv-import-survey.md) |\n| `Q-c2prime-drift` | PARTIAL | Why does c2' drift on the 22-term ladder? | Class-count-blind (attack-c2drift-01 §1); on item 1c's own band x >= 41 the ten-point rate is unresolvable: c2' 0.3745 +/- 0.1411 against c1's 0.3344 +/- 0.2358 on the same primes, a difference of 0.0401 = 0.12 of the control's ten-point window sd. Whether the drift is real is NOT closed here: attack-c2drift-01 measured it at p = 0.0053 on x >= 17 and left (i)' and (ii) alive. exponent-control §2/§4/§6 refit at 22 terms. | 1c (retired) | [attack-c2drift-01.md](history/staging/attack-c2drift-01.md), [c2prime-refit-22.md](history/staging/c2prime-refit-22.md) |\n| `Q-cal4-routing` | ANSWERED | Does the standing certificate plan point at the right obstruction, and what is CAL4 worth at @17? | The plan is aimed at the wrong obstruction, but not for partition W's reason: the @13 1e-9 gate is met today only by an assembly containing the brute-force C(990,4) march whose @17 counterpart is 513 days on ten cores and is not codeable here, the @17 requirement is a ladder whose first useful rung is relT4 <= 5.05e-5, a factor of 7.9 rather than six orders, and CAL4 has no correct constant value. | none | [phase1-W2-cal4-routing.md](history/staging/phase1-W2-cal4-routing.md) |\n| `Q-calm-lemma-identities` | PARTIAL | What in the fused window and the anchored calm is proven, exactly? | Three identities and the exact spectral forms are proven and verified as exact integers or floats; the uniform-in-q anticorrelation is REFUTED rather than open, with six counterexample primes across six levels, and the Anchored Typicality Measurement is the anchored-escape wall one level down with no proof mechanism in sight. | none | [natal-cap-19-calm-lemma.md](natal-cap-19-calm-lemma.md) |\n| `Q-calm-vs-kill` | PARTIAL | What does the anchored calm buy against annihilation? | Two theorems, the @11 closure and the X-Limitation Theorem at @11, @13, @17 and @19, and two refutations, calm implying concentration and minimizer loudness; the overlap-floor candidate that would prove non-annihilation dies on the anchor's own drift, with the anchored rotation t = 0 its worst case at @17, leaving the X-floor open. | none | [natal-cap-31-calm-vs-kill.md](natal-cap-31-calm-vs-kill.md) |\n| `Q-capK-bv` | ANSWERED | Is the prime-regime cap_K over the full wheel an unconditional asymptotic via the fundamental lemma over Bombieri-Vinogradov? | PROVEN at short-note grade for q_K = W^{o(1)}, and EMPTY in every computable range (needs s >= 10.82, first cleared at x = 263; the lemma's floor s >= 10 is first reached at x = 239); the deep ladder is untouched. | 11 (retired) | [thm-capK-bv.md](history/staging/thm-capK-bv.md) |\n| `Q-capture-identity` | ANSWERED | Does the transplanted cap family collapse onto a classical object, and what does the collapse say about the depth law and the wall? | PROVEN, elementary: floor_K = T − X(K) on the half-open window with both members in [Q², Q′²), a convention that is load-bearing; the wall is renamed, not weakened; Σ capU counts each member against its own threshold, not against a uniform p_K, the two differing 11% to 37% at K*. | Z2 | [quadpoint-identity-01.md](history/staging/quadpoint-identity-01.md) |\n| `Q-centered-discrepancy-estimate` | PARTIAL | Can the top of D_y be removed using ordinary prime distribution after an exact divisor flip, and what estimate remains? | Equations (3)-(4), the exact split and flip, survive review and finite controls. A repair of the top-range bound T^top=O_(A,eps)(x/log^A x) is DERIVED in section 3a (2026-09-08) with the endpoint atom retained, odd square divisors, the corrected reciprocal-totient local factors with (m',g)=1, explicit truncation powers, a Cauchy multiplicity device, and a uniform Mobius mean proved from the q=1 Siegel-Walfisz statements; its inputs are the terminal-point BV of Tao Notes 3 Theorem 17 with the prefix form derived by rounding. The handler (lane V) read the derivation on 2026-09-08, reconstructed (BV*), (3a.9), (3a.14) and (3a.16), checked the limiting constant of (3a.9) numerically at eight (b,g) pairs and found no defect; it is accepted at its stated scope, with ineffective constants. The payoff is D_y=D^(e_1)+O_(A,eps)(x/log^A x) only, so the handoff consumer is equivalent to D^(e_1)>=-4x/25+o(x). The fixed-endpoint signed discrepancy, D_y>=-4x/25+o(x) and twin-prime infinitude remain OPEN. | C | [centered-discrepancy-estimate.md](centered-discrepancy-estimate.md) |\n| `Q-centered-discrepancy-measurement` | ANSWERED | At reachable x, is the OPEN sufficient input D_y>=-4x/25 numerically violated, is the absolute form (13) numerically plausible, and does D_y carry structure beyond a random-sign model? | MEASURED to x=2^38. Neither pre-registered falsifier fires. D_y/x stays within 0.0043 of zero for the measured j>=26. Over 30<=j<=38, D_y/x differs from -(T1/x-C2) by at most 2e-4, with maximum 1.84401e-4 at j=30; the classical term dominates this finite comparison. The data do not establish an asymptotic fluctuation scale, an impossibility of informative future measurements, or the sufficient signed estimate. No proof status changes. | C | [centered-discrepancy-measurement.md](centered-discrepancy-measurement.md) |\n| `Q-certificate-engine` | PARTIAL | What did natal-cap-28 assemble, and at what calibration does the certificate engine stand? | There is no fully analytic certificate law and nothing here is closed: what reads as one object is nine at five calibrations with one of the nine refuted, the Certified-Head Theorem is PROVEN on 1/3/6 certified primes at @17/@19/@23, and every engine output carries the status of its weakest input, the HEURISTIC Buchstab Transfer Hypothesis. | none | [certificate-engine.md](certificate-engine.md) |\n| `Q-chain-review-0906` | ANSWERED | Do the written derivations from S(x) down to E_dagger hold as stated, with every imported theorem used inside its hypotheses and every discarded term at the claimed O_H(x/log^H x) rate, and where does the chain remain unreviewed? | Three independent adversarial passes found no defect in the six classical contributions from S(x) to E_>, none in the grouped moment (1)-(20) including the E_* to E_dagger step, and none in the residual-coverage link E_> to E_*, whose block-shape derivation was reconstructed and verified exactly. Every imported theorem was confirmed in its primary source except the Mobius Bombieri-Vinogradov citation, which points at an exercise sheet and needs a published replacement. Stale o(x) rates in three note bodies were replaced. A checked argument is not a refereed theorem; the OPEN signed margin is untouched. | C | [chain-review-0906.md](chain-review-0906.md) |\n| `Q-changelog-add-B` | ANSWERED | Which superseded text from wave-2 partition B belongs in the changelog? | The per-document entries for U-FRAME, ATTACKS3 and gate-multiplies, staged for the shepherd to merge in one pass; nothing was written to CHANGELOG.md here. | none | [changelog-add-B.md](history/staging/changelog-add-B.md) |\n| `Q-changelog-add-C` | ANSWERED | What history left wave-2 partition C's paper-suite files, and what has to be staged for the CHANGELOG? | Staged: Door 2's two load-bearing numbers are RETRACTED because the column they came from was disclaimed by its own script's correction banner, the corrected artifact reverses the conclusion, the OEIS term count is corrected to twelve, the matching lower bound is retired, and Zone Equivalence is demoted from spine theorem to framing device. | none | [changelog-add-C.md](history/staging/changelog-add-C.md) |\n| `Q-changelog-add-F` | ANSWERED | Which superseded text from wave-2 partition F, the natal-cap family, belongs in the changelog? | Entries for eleven natal-cap documents and two new files, led by the retirement of the Fused-Window Calm Lemma and the certificate law with it, plus the prose-note index that closes qc-arch E4; staged only, not merged. | none | [changelog-add-F.md](history/staging/changelog-add-F.md) |\n| `Q-changelog-add-G` | ANSWERED | What history left wave-3 partition G's files, and what has to be staged for the CHANGELOG? | Staged verbatim with file and line: the twin-candidate census index is corrected from A059861(n-1) to A059861(n) with four independent confirmations, two script banner titles are retired as contradicted by their own companion prose, and two scripts that had no banner title gained one. | none | [changelog-add-G.md](history/staging/changelog-add-G.md) |\n| `Q-changelog-add-P` | ANSWERED | Which superseded text from wave-4 partition P's three files belongs in the changelog, and what does calibration against the home document find? | Fourteen defects came out of the check wave 3 recorded it had not run, and none of them is reachable by any existing gate check; the removed or rewritten text is reproduced verbatim per file for the shepherd. | none | [changelog-add-P.md](history/staging/changelog-add-P.md) |\n| `Q-changelog-add-Q` | ANSWERED | What history left wave-4 partition Q's files, and what has to be staged for the CHANGELOG? | Staged: the Maier-Pomerance clause in the OEIS draft is RETIRED because it drew a >= from a << (they state an upper bound), and audit-numbers.js gained an oeis part of 39 checks carrying three retired values, including the exponent 1.57 quoted for G2 when 1.57 is A288815's. | none | [changelog-add-Q.md](history/staging/changelog-add-Q.md) |\n| `Q-changelog-wave2-A` | ANSWERED | What changelog entries does wave-2 partition A owe research/history/CHANGELOG.md? | Staged per-document entries for G2-STATE.md, LOCALIZED-GAP.md, maier-matrix.md and origin-excess.md, each naming the claim as it stood, what replaced it and the artifact that forced the change; a staging file for the parent to merge, not an edit to the changelog. | none | [changelog-add-A.md](history/staging/changelog-add-A.md) |\n| `Q-changelog-wave2-D` | ANSWERED | Which changelog entries does wave 2 partition D stage for CHANGELOG.md? | Staged entries in the changelog's house form for research/README.md (repurposed from a summary into the router), README.md, GLOSSARY.md, TODO.md, THE-DIALS.md, ZONE-POSTULATE.md, THE-LENS.md and FOLD-PROFILE.md, for the parent to merge. | none | [changelog-add-D.md](history/staging/changelog-add-D.md) |\n| `Q-changelog-wave2-E` | ANSWERED | What changelog entries does wave-2 partition E owe research/history/CHANGELOG.md? | Staged entries covering covering-dive.md, two-class-lower-bounds.md, PRIOR-ART.md, sift-limit-attack.md, theta-ladder.md, maxgap-law.md and dhr-verification.md, led by the FKMPT constant moved across two files together. | none | [changelog-add-E.md](history/staging/changelog-add-E.md) |\n| `Q-changelog-wave3-H` | ANSWERED | Which changelog entries does wave 3 partition H stage for CHANGELOG.md? | Entries for U-FRAME.md's split, in which sections 10, 11 and 12 keep their numbers and heading text and lose their bodies, plus the three new child documents kappa-not-L.md, operator-and-pair-count.md and f-decays.md. | none | [changelog-add-H.md](history/staging/changelog-add-H.md) |\n| `Q-changelog-wave3-I` | ANSWERED | What changelog entries does wave-3 partition I, the flagship split, owe research/history/CHANGELOG.md? | Staged entries for the split of the flagship's wall chapter into the new paper/wall-note.md and the pointer repairs that split forces, in house form, newest first and grouped by the document changed. | none | [changelog-add-I.md](history/staging/changelog-add-I.md) |\n| `Q-chen-fold-benchmark` | ANSWERED | Can the fold framework reproduce a classical Chen lower bound on long intervals with every imported hypothesis and error budget stated? | The classical implication is derived below using three explicitly imported theorem statements, with the combinatorial minorant proved and its integral margin certified by exact rational arithmetic. This is a benchmark, not a new prime theorem, and it does not prove twins or a short-zone result. | C | [chen-fold-benchmark.md](chen-fold-benchmark.md) |\n| `Q-chen-opportunity` | ANSWERED | Does the signed Chen target identify an approachable proof mechanism, and what should the next research obligation be? | The target is sufficient but subtracts an avoidable odd-composite penalty; even cancellation of its obvious signed components does not close the existing separate bounds. Select a prime-detecting weight against explicitly justified arithmetic input ranges before pursuing a new estimate. The opportunity is a research judgment, not a proved mechanism. | C | [chen-opportunity-audit.md](chen-opportunity-audit.md) |\n| `Q-chen-signed-target` | PARTIAL | What exact additional arithmetic estimate would turn the completed Chen benchmark into a proof of infinitely many twin primes? | An elementary prime minorant gives a conditional twin theorem with all weights, ranges, quantifiers and errors specified. The opportunity audit identifies an avoidable negative-weight penalty and a separate-estimate budget that fails even under component cancellation; this sufficient target remains OPEN and is not the sole next proof obligation. | C | [chen-signed-target.md](chen-signed-target.md) |\n| `Q-coefficient-structure` | ANSWERED | Does the arithmetic structure of the aggregated coefficients reduce the endpoint estimate beyond the previous rectangle, and can smaller coefficient norms alone supply the remaining power saving? | Exact classification gives a nonsquarefree L2 norm of size at most sqrt(D)*W^(1/4) times logarithms, while the squarefree top-band energy is at least a constant times D*W*log(W). At d~x^0.277, e~x^0.467 every endpoint sector except the squarefree-squarefree prime sum has a power saving. This reduces that rectangle to one explicit endpoint correlation; prime-dispersion.md supplies the remaining estimate and controls the full rectangle. Finite algebra, CRT sectors and archived prefixes are checked separately. | C | [coefficient-structure.md](coefficient-structure.md) |\n| `Q-cofactor-progression-transfer` | PARTIAL | Can the corrected full-cofactor kernel be estimated on a growing cofactor range by representing its actual arithmetic on CRT progressions, and does the resulting bound reach the twin consumer? | Derived from named imports: for some positive absolute kappa and d, all prime-r cofactor pairs s,t <= floor((log x)^kappa), including non-squarefree inputs and mixed transition/smoothed profiles, admit a dyadic scale-average O(log^(2-d) X) bound after division by x. The extension uses bounded multiplicative Euler modifications on the progression s\\|n, t\\|n-2 and keeps its density 1/lcm(s,t), so summing cofactors costs only O((log log X)^2). This goes beyond s=t=1 but remains short of o(x). The large-cofactor joint remainder, outside residual and twin margin remain OPEN; the attempted extension to power-sized cofactors fails on source modulus range and accumulated logarithmic cost. | C | [cofactor-progression-transfer.md](cofactor-progression-transfer.md) |\n| `Q-coherence-0828` | ANSWERED | Are the 2026-08-28 notes consistent with each other? | 20 contradictions, 8 convergences, 46 live-doc corrections collected (34 mechanical, 12 need a second reader). | none | [coherence-0828.md](history/staging/coherence-0828.md) |\n| `Q-coherence-0829` | ANSWERED | After the 2026-08-29 edits, does every live document state each changed fact the same way, and do the papers still agree with the notes? | 24 changed facts checked at 63 live-layer sites: 16 live disagreements (11 APPLY, 5 HOLD) plus one count error inside a HELD note; the worst is the localized band 2.9 to 4.2 attributed to \"two engines\" when only one of the two reads it, the other reading 2.14 to 4.39; qc.js 1 finding, in a file another agent holds; audit-numbers 251/251. | none | [coherence-0829.md](history/staging/coherence-0829.md) |\n| `Q-coherence-0830` | ANSWERED | Do the live layer (TODO, README, G2-STATE, REFUTED, anchored-calm, GLOSSARY, CHANGELOG, QUESTIONS) and the eighteen notes landed 2026-08-29 evening and 2026-08-30 morning agree sentence by sentence? | They mostly agree; 27 distinct disagreements found, 15 graded APPLY (both sides quoted, no judgement needed) and 12 ASK, none touching a note's mathematics: the live layer lags the landed notes in six places (a PROPOSAL listed under PROVEN in G2-STATE section 0, 67 for 71 closed routes, the skeleton door still open in README and two anchored-calm sentences, Z5's body contradicting the record-mechanism note, item 9's one-versus-two open steps, a stale Z2 first move), carries four numeric slips (8-of-10 for 10-of-10, 27 for 25 clauses, u* in the seventh digit, nine for ten levels), one rung inflation (a text-layer reading called read at the page), and one landed note (recon-0830-rec-killrun) is itself wrong about the record on the seventy-five count; qc.js --full at 09:23 fails only on two ledger ids from notes that landed during the run. | none | [coherence-0830.md](history/staging/coherence-0830.md) |\n| `Q-comb-discrepancy` | PARTIAL | Can the Comb Discrepancy Lemma's constant 2*3^k be tightened (TODO 8c)? | The lemma IS the trivial per-block bound; the attained optimum is max G - min G, a computed number per level (slack 29x @23, 54x @29, growing); term-by-term pricing lifts the certified head share to about 28-29 percent, not the 50 hoped for. | 8 | [comb-discrepancy-tight.md](history/staging/comb-discrepancy-tight.md) |\n| `Q-comb-tail-0830` | PARTIAL | Can TODO item 8's remaining part (b), prime-comb equidistribution in the tail regime, be proven at the depths the certificate needs, or at least measured and priced at the enumerable levels (8b)? | Not proven and no exponent moves: the tail defect is measured exactly for the first time (signed -0.01% of the tail main term at @23, -0.00% at @29, unsigned 0.60% and 0.25%, smaller than the head's 1.00% in aggregate but 8 to 37% per prime at its worst), the Comb Discrepancy Lemma's method fails in the tail on its MAIN TERM (off by a derived factor e^gamma(1/u - q^(1-u)), measured aggregate 1.14x at @23) before its error term (10^27 x main), the prime-comb Legendre form has 2^(k+2) terms of which the short ones obey the lemma's <1 bound and the long ones are prime-in-AP discrepancies whose term-by-term sum is 11.2% of main at @23 against a signed -0.01%; the all-x form at K = 0 is Theorem C (PROVEN, limit only) and at the certificate's depth K* the non-closing step is a kappa = 1 asymptotic at sifting parameter s < 2, a proof gap on data that show B_tail = 0.998-1.000 at K* through @29; the full-depth half is 8(a), CLOSED. | 8 | [attack-0830-comb-tail.md](history/staging/attack-0830-comb-tail.md) |\n| `Q-consolidation-wave` | ANSWERED | Can the five cost-class scripts be bound to their own output, and can U-FRAME's two lost tables be rebuilt? | All four runnable cost-class embeds are bound and the hand-pasted-tail counter falls 19 to 15 with the gate at TOTAL 0; three needed --force at a cost of 24 tokens out of 211 with no value changed anywhere, and natal-cap-33-overnight.js is a three-run composite that should never have been on the cost list. | 11b (retired) | [consolidation-wave.md](history/staging/consolidation-wave.md) |\n| `Q-constants-audit-2` | ANSWERED | For every load-bearing number that appears in two or more live documents, does it agree everywhere, and does it trace to exactly one custody-bound producer? | 25 findings over 69 of 70 files, 8 rank A, 12 rank B, 5 rank C: 46 named-constant groups carry two or more distinct values across two or more files, and 439 numerals of five or more significant digits in the live layer have no producer printing them at any precision; nothing was fixed and five judgement calls stayed queued. | none | [constants-audit-2.md](history/staging/constants-audit-2.md) |\n| `Q-consumer-comparison` | ANSWERED | How does the sufficient margin C_2 x + E_dagger(x) >= c_0 x/(log x)^K compare logically with the published sufficient routes to twin primes, and what does the literature already own of the corner object? | The sufficient margin remains OPEN. The published conditional routes imply it where their hypotheses and stated lower bounds apply; reverse implications are not established. The dictionary is E_dagger(x)/x = C_2(2 delta_x - delta_{x/2} - 1) + o(1), with K>0 below that normalisation's resolution. Independent review corrects the comparison: cumulative and dyadic consumers on unbounded scales are equivalent at the same fixed K with possibly different positive constants. At K=0 an eventual cumulative lower bound also implies the unbounded-scale dyadic consumer. The nonnegative-weight corner identification applies only to the prime-cofactor subfamily; the full residual retains additional branches. Every surveyed route remains conditional. | C | [consumer-comparison.md](consumer-comparison.md) |\n| `Q-corner-branch-diagnostic` | ANSWERED | Can future experiments enumerate the full coefficients C(n) and C'(n-2) branch by branch, without silently dropping proper prime powers, cofactors s>1, non-squarefree inputs or either gcd branch? | The driver exists and its two independent constructions of each side agree exactly on 8192 retained sides and 1802 fixture sides; five deletion and reconstruction controls are ACTIVE in both runs. Its split by gcd(n,n-2) measures integer parity, not the per-term CRT gcd partition. The fixed-eta corner is empty on the retained prefix, and Elo exceeds E0 across the entire J_x, so the matrix is a proxy window. This is finite algebra and tool preparation, with no asymptotic estimate or twin margin. | C | [corner-branch-diagnostic.md](corner-branch-diagnostic.md) |\n| `Q-corner-coefficient-energy` | ANSWERED | Can the full corner coefficients be studied before splitting their cofactor branches, and what do exact energy and classical smoothed sieve weights supply? | The sharp coefficient energy is an explicit signed quadratic form in short cofactors plus O_epsilon(x^(1/2)K^2*x^epsilon), a power below x at the fixed corner. Logarithmically averaging the divisor cutoff gives Barban–Vehov weights and an O_eta(x log x) squared norm on each side and absolute shifted-product bound. The follow-up sharp-corner-transition.md gives sharp O_eta(x log^2 x) norms and proves a non-negligible norm transition for sufficiently small fixed eta. Signed transition saving and the global complement remain OPEN; no exact residual cut or twin margin changes. | C | [corner-coefficient-energy.md](corner-coefficient-energy.md) |\n| `Q-corner-correlation` | PARTIAL | On the corner S_0 where both cofactors sit just above their own cutoffs, what exactly is the remaining endpoint sum, what does the sufficient consumer need there, and does any reviewed correlation theorem supply it? | The full corner is exactly sum_n C(n)C'(n-2). Proper-prime-power terms above the cutoffs are negligible at fixed eta by §1.1; the remaining prime-r terms with s>1 or s'>1 and non-squarefree inputs remain OPEN. The s=s'=1 piece has a nonnegative Mobius-product weight with exact moving cuts. The multiplicative lift and round-review sampling give small continuous and dyadic scale-average log savings for that subfamily, with no o(x), full-corner control or twin margin. The complement is still open. | C | [corner-correlation.md](corner-correlation.md) |\n| `Q-corner-log-average` | PARTIAL | What do logarithmically averaged correlation theorems and their quantitative successors actually supply for the prime-cofactor corner weight, with exact support, rates and scale quantifiers? | The pure band is invariant under dilation by primes outside it; the exact cofactor window is not, but its support extends to x^(1-2eta) and need not leave the window under a fixed dilation. Uniform-prefix logarithmic bounds imply block bounds by Abel summation; a relative exponent above 3 is one sufficient route to o(x), not a universal necessary threshold. Pilatte's 1/(96e) calculation applies only to his displayed parameter choice. A bounded multiplicative Fourier lift now gives a small continuous scale-average log saving for the prime-cofactor subfamily via Tao-Teravainen v2; see prime-band-transfer.md. No o(x), dyadic pointwise estimate, full-corner estimate or positive twin margin follows. | C | [corner-log-average.md](corner-log-average.md) |\n| `Q-corner-measurement` | PARTIAL | At finite dyadic x, how large is the corner two-point correlation K(x)=sum_{x/2= 1 with the density condition as its only hypothesis, so what beta2-note adds is two parameter choices and one dimension-free observation that is zeb's at kappa = 1; the honest response is to state the k-class family, whose k = 1 row is a known theorem. | none | [attack-dim2-standalone.md](history/staging/attack-dim2-standalone.md) |\n| `Q-discrepancy-two-class` | ANSWERED | What is the two-class analogue of Holt's signed discrepancy DeltaPhi, and do the three candidate growth laws on the table agree? | One object in three norms rather than three agreeing measurements: the proven Level Ledger bound 2*3^(pi(x)-1) and natal-cap-29's spectral level law share the integer 3 provably, since R_k(p) tends to k+1, so the apparent agreement is an identity; the custody gate reproducing Holt's Table 2 passed. | none | [discrepancy-two-class.md](discrepancy-two-class.md) |\n| `Q-dispersion-range` | ANSWERED | How far do the dispersion and coefficient-sector estimates extend across divisor sizes, and does a better orientation or gcd budget enlarge their overlap? | Averaging the nonzero gcd losses over the original divisor gives a different dispersion bound with cross-prime exponent (5/4)a+(1/2)b+3/50, where a=delta+6/25 and b=nu+1/20. Explicit diagonal, prime-cut and sparse-sector inequalities control a region of full rectangles, including d~x^0.282, e~x^0.522 at product scale x^0.804. Its cross-prime and left-sparse budgets are 0.9985 and 0.9988. The supremum of product exponents in this sufficient region is 5397/6700; it is not a uniform product cutoff. Balanced supports separately allow 12/25= pi(2s)-pi(s) (K* = 17 at s = 16 by exact walk, VERIFIED, so the certificate reads 18 against 8 on the chain itself; it exits the whole legal band at s = 128, PROVEN); the sharper maxsum bridge Ghat(2s) <= maxsum_{K*+1}(T_s) is PROVEN and holds under 8 at all fourteen enumerable steps (VERIFIED), but its all-s form needs an upper bound on K* against the entering primes' two-class covering that nothing proven supplies; the doubling inequality itself is untouched. | D | [attack-0829n-doubling-bridge.md](history/staging/attack-0829n-doubling-bridge.md) |\n| `Q-doubling-killrun-0830` | PARTIAL | Can the exact per-fold L bounds of the pi(2s) - pi(s) folds inside one doubling step compose to an upper bound on the weighted kill-run below the allowance 8 Ghat(s)/gbar(s), for all s, without passing through K*? | No, and the composition is closed as a route, not merely unproven: the folds compose as a PRODUCT, K*+1 <= prod_j (1+L_j) (PROVEN here, exact at fourteen steps, 540 against 18 at s = 16), the composed certificate never sits below yesterday's maxsum certificate (PROVEN by monotonicity), it exceeds the allowance at 8 of 14 enumerable steps starting at s = 9 (VERIFIED), and since every fold kills a slot the composed index is at least 2^N, whose floor 2^N gbar(s) alone exceeds 8 Ghat(s) at the chain rungs 16, 32, 64 and beats the cited polynomial ceiling on Ghat for all large s (PROVEN given the ceiling); the sum form is false at seven steps and the max form is a floor; the weighted run itself, computed exactly, sits at 0.83 of its allowance at s = 16 and (M8) is exactly where yesterday left it, OPEN. | D | [attack-0830-doubling-killrun.md](history/staging/attack-0830-doubling-killrun.md) |\n| `Q-embed-backlog-0829` | PARTIAL | How much of the embed backlog (legacy scripts with no OUTPUT block, readings not traceable to their block) could be closed in one pass, and what did the fresh runs contradict? | MEASURED. embed-backlog fell 218 -> 189 findings in one pass: no-output-block 34 -> 5, readings-not-traceable 181 -> 181 (untouched by design), hand-pasted-tail 3 -> 3. Twenty-nine legacy scripts were bound with node research/qc/embed.js, zero --force was used, and one legacy reading is contradicted by its own fresh output (attack-beta2-03-exact-strata reading 8, \"1.04 to 1.07 across y = 7..23\" against a printed 1.102 at y = 7). | none | [embed-backlog-0829.md](history/staging/embed-backlog-0829.md) |\n| `Q-embed-binding` | ANSWERED | What does the embed fingerprint actually bind, and what does it leave loose? | It binds the code to a claim about the output and does not bind the output block to anything: code-sha256 covers every byte above the banner, out-sha256 is only ever compared against another run, timeouts and failures write nothing, a prose line can unbind a tail and one did during this pass, and 37 cited scripts have no output custody at all. | none | [verify-the-verifier-embeds.md](history/staging/verify-the-verifier-embeds.md) |\n| `Q-endpoint-fourier` | ANSWERED | Does a complete Fourier truncation budget and a matched Kloosterman-fraction theorem control any further part of the actual endpoint sum when its small prime-power coefficients are combined first? | A written classical-input derivation controls the rectangle d in (floor(x^(27/100)),2floor(x^(27/100))], e in (floor(x^(46/100)),2floor(x^(46/100))] to O_H(x/log^H x) for every fixed H. Its product is of order x^(73/100), outside the previously controlled region eventually. The aggregated Fourier exponent is 1989/2000; the full Vaaler tail and gcd=2 branch are included. This is a regional estimate, not a full residual bound or twin theorem; finite algebra and saved-factor checks are separate validation. | C | [endpoint-fourier.md](endpoint-fourier.md) |\n| `Q-endpoint-pairing` | ANSWERED | Does retaining the difference of interval endpoints improve the complete Fourier budget, and are the lowest frequencies actually the next obstruction on the squarefree pilot rectangle? | A written derivation controls the full rectangle d~x^(11/40), e~x^(93/200) to O_H(x/log^H x), with product of order x^(37/50). The sufficient first exponent becomes 3/20+(7/10)(a+b)+(1/4)max(a,b), equal to 3999/4000 there. On the a=b=517/1000 pilot, this transition-frequency budget is 20061/20000, not the separate-endpoint 20163/20000; prime-dispersion.md now controls that pilot with a different estimate. Finite identities and exact exponent checks validate the implementation, not the asymptotic theorem. No uniform product cutoff or twin lower bound follows. | C | [endpoint-pairing.md](endpoint-pairing.md) |\n| `Q-endpoint-target-audit` | ANSWERED | Does the endpoint reduction require a fixed positive fraction of the expected twin count, and do recorded uniform-gap theorem failures exclude its weaker consumer? | No fixed fraction is required: for every fixed H the reduction has error O_H(x/log^H x), so C2*x+E_>(x)>=c*x/log^K x on unbounded dyadic scales suffices, as does a stated logarithmically rescaled average. These implications are derived from named inputs; their endpoint hypotheses remain OPEN. The recorded uniform-gap theorem comparison has different quantifiers and does not establish an obstruction for this consumer. | C | [endpoint-target-audit.md](endpoint-target-audit.md) |\n| `Q-erdos-map` | ANSWERED | Does the two-class object already occupy a numbered slot on somebody else's map of open problems? | It has no problem number and no name on erdosproblems.com, Ben Green's list or the OEIS citation graph, but it is not unposed: Erdos posed it himself in one sentence in the same paragraph as the $1000 offer, and our object is the r = 2, difference-2 instance of his own stated extension of #687. | none | [attack-erdos-map.md](history/staging/attack-erdos-map.md) |\n| `Q-excess-chain-c` | CLOSED | Does the excess chain's constant c pin to 1.074 via a compressed-tail max correction? | NO: for any FIXED correlation length the derived family tends to 1 and both named corrections lower the maximum, so 1.074 is not derivable from a compressed-tail max correction; c measures 1.05 +/- 0.07 (sd) over x = 17..31, MEASURED, which does not separate 1.05 from 1.074. | 10 (retired) | [excess-chain-c.md](history/staging/excess-chain-c.md) |\n| `Q-exponent-control` | PARTIAL | What is the growth exponent of G2, measured against a control that knows its own answer? | MEASURED 1.50 +/- 0.05 stat on 22 trusted terms after control correction, systematic unquantified; the raw fit's bias grows and sticks at +0.28, exponent 2 stays disfavoured but is not excluded, and the question is bias-limited rather than data-limited. | none | [exponent-control.md](exponent-control.md) |\n| `Q-external-data-audit` | ANSWERED | Where is the corpus limited by external data it could simply adopt? | Findings are candidates, not verdicts: the highest settles an open reading for free, since dividing the two now-trusted series moves the G2/h ladder maximum from x = 37 to x = 79 and reverses the fall at x = 53, killing \"peaks at x = 37\"; the rest tighten quoted constants or retire stale limitation sentences. | none | [external-data-audit.md](history/staging/external-data-audit.md) |\n| `Q-f-census` | PARTIAL | What does the grain census law give for f, the qualifying-gap fraction, and where does it structurally stop? | DEFECT-carrying: every census point from x = 37 up is aliased, so the 42-point table above x = 31, both regressions, the factor 170 and the x = 1000 extrapolation must be re-read against the corrected law; the staircase theorem (treads are exactly the twin pairs), Lemma B, the comb coefficient and everything at x <= 31 are UNAFFECTED, and the growth law is a fit and not a theorem. | none | [a3-03-f-from-census.md](a3-03-f-from-census.md), [f-decays.md](f-decays.md) |\n| `Q-f4weak` | PARTIAL | What is the weakest sufficient form of F4 that could be proven, given rms <= z^(3+o(1))? | The mode-count/degree bound is NOT the theorem (sqrt(N)*rms grows at z^8.174 +/- 0.249 against the z^4.26645 ceiling and is dead from z = 31); one new PROVEN finite-z clearing survives z = 47 at a 5.0 percent margin but converges onto the trivial bound; the weakest law-shaped form found is MV(alpha), a mean value over modes, MEASURED slope 4.214 +/- 0.304 with beta2 inside the bar. | none | [attack-f4weak-01.md](history/staging/attack-f4weak-01.md) |\n| `Q-fdecay-deep` | ANSWERED | Does the fitted f decay law hold out of sample at deep levels? | It holds as a band, nine of nine deep levels inside the four-specification band, but the residual trends against the central specification at t = -7.08; the test also found what it was not looking for, that the 42-point exact census is wrong from x = 37 upward by a factor rising to 1.63 at x = 199, from a 32-bit shift alias. | none | [fdecay-deep.md](history/staging/fdecay-deep.md) |\n| `Q-fdecay-out-of-sample` | OPEN | Does the 42-point decay law for f, the qualifying-gap fraction, survive out of sample, given that every downstream reading extrapolates it three to thirty times past its range? | Pre-registration only, written before any producer for the pass exists on disk: four specifications are refit on the 42 exact census points as a transcription check and their projections are frozen, and nothing here is measured. | none | [fdecay-deep-prereg.md](history/staging/fdecay-deep-prereg.md) |\n| `Q-fekete-1d-defect47` | PARTIAL | Does the bounded-defect Fekete route survive a probe at 47#? | The defect at 47# is ordinary and the 47# enumeration cannot move item 1d's TPC threshold whatever value it returns; the bounded-defect Fekete lemma is stated exactly and proved with every hypothesis except the candidate itself discharged, so the route reduces to one named inequality and is neither closed nor open beyond that. | 1d | [fekete-1d.md](history/staging/fekete-1d.md) |\n| `Q-fixed-endpoint-discrepancy` | PARTIAL | After the accepted truncation to odd moduli e=-4x/25+o(x)? | Reviewed 2026-09-09 after the coprimality repair: T_I^low=O_(A,eps')(x/log^A x), hence S=C_2x+B+O_A(x/log^A x), with B the exact Type II plus band sum (2.9). The untruncated modulus bound was false; g<=(log x)^(A+13), both tails, the coprime density and the weighted BV multiplicity now pay the estimate. Review corrects harmless log powers and the power-of-two atom over the full eps' range. D^(e_1)>=-4x/25+o(x) is equivalent to B+2C_2M>=-4x/25+o(x); it implies, but is not equivalent to, B>=-(C_2-1/200)x+o(x). The latter and H_B are sufficient OPEN margins. The absolute band statement (4.9) is one stronger sufficient input, not a necessary condition. No inspected source estimates the remaining actual signed B. Twin-prime infinitude remains OPEN. | C | [fixed-endpoint-discrepancy.md](fixed-endpoint-discrepancy.md) |\n| `Q-fkmpt-corrigendum` | ANSWERED | Does the 2023 FKMPT corrigendum change anything the corpus depends on? | Clean bill of health: the corrigendum changes four numerical constants and the parameter M, every one already carried at its corrected value here; the loudest finding is the opposite of the expected one, since the premise that nothing in this repository has read it is false and has been since 2026-08-18. | none | [verify-fkmpt-corrigendum.md](history/staging/verify-fkmpt-corrigendum.md) |\n| `Q-fold-arithmetic-bridge` | PARTIAL | Does one parity-table bridge from the anchored fold ledger yield a sufficient twin lower bound with named arithmetic inputs? | Exact identities retained. The one unread sieve input of the pricing, Bombieri--Vinogradov for k-fold X^(1/u)-rough products, is derived from Wu's Lemma 2.3 (section 3a); the same input, sieved in the composite variable, gives the contamination aggregate constant 4 for each fixed k,u, replacing the displayed 12.86 to 19.72. With these inputs, elementary bounds give Q_cov(u)<1 and c*_real(u)<4 for every u>4 (section 4a, independently reviewed 2026-09-09 with rational certificates), so neither sufficient ratio test succeeds at any depth. This closes the two tests, not the decorrelation hypotheses, and supplies no twin estimate. | C | [fold-arithmetic-bridge.md](fold-arithmetic-bridge.md) |\n| `Q-fold-profile` | ANSWERED | Where does a fold by p land its damage inside the original tile stretch [0, W)? | The counting half is a theorem with an effective constant (Level Ledger PROVEN, Mirror Ledger VERIFIED 45 of 45) and the survival law is scale free (MEASURED), but the quantity controlled here is provably uninformative about L and kappa(m), so it buys nothing against the Zone Postulate route. | none | [FOLD-PROFILE.md](FOLD-PROFILE.md) |\n| `Q-fold-succession` | CLOSED | Does folding in succession damp the multiplier, giving a lower L than the sum of per-fold jumps? | No: the succession-damping route closes empirically at both reachable scales against a calibrated null, the only damping present is what level-tracking forces and the only extra structure present is cooperative, matching the structural closure by super-additivity. | none | [fold-succession-autocorr.md](history/staging/fold-succession-autocorr.md) |\n| `Q-foldL-amortized` | CLOSED | Can the fold multiplier L be bounded by charging kill-runs to the birth of the gaps they consume? | The accounting does not close on the tile: the Consumption Identity makes the charge exact and the bound is then only as good as X = 2Q0 + Q+ + Q-, and supply against demand fails on the tile at every x by exactly the factor ln^2 x; it closes in a localized window from p about 421 on, which is a statement about a bounded population only. | none | [attack-foldL-04-amortized.md](history/staging/attack-foldL-04-amortized.md) |\n| `Q-foldL-bridge` | ANSWERED | Does the bridge from a fold's kill count to L hold, and what is the sharpest transportable ceiling it gives? | The bridge is PROVEN and was already in the corpus, so the job was custody rather than a new theorem; the new pair-window rung is exact at all eight diagonal cells, and the best fully-proven transportable ceiling stays near 0.18p and beats the requirement nowhere. | none | [attack-foldL-02-bridge.md](history/staging/attack-foldL-02-bridge.md) |\n| `Q-foldL-extinction-scaling` | ANSWERED | Does the multi-kill extinction law predict windows it was not fitted on? | CONFIRMED under the pre-registered rule at Y = 2e7, 2e8 and 2e10 from the Y = 2e9 calibration alone, with the run-length extension REFUTED exactly as pre-registered; this promotes a law candidate on measurement, not a proof. | none | [attack-foldL-06-scaling.md](history/staging/attack-foldL-06-scaling.md) |\n| `Q-foldL-maxsum-direct` | CLOSED | Can maxsum_m(T_x), plain and residue-deleted, be bounded above by an argument that never counts kills? | The brief's premise contradicts the corpus's own Bridge Floor (maxsum_k >= maxsum_1 = G2(T_x) for every k), so the bridge cannot convert far enough however good the bound is; all four routes close (R1 partial then closed, R2 tautological and quantitative, R3 busts the budget), and the order-m object is in print, so item 0c's absence sentence has to change. | 0c | [attack-foldL-05-maxsum-direct.md](history/staging/attack-foldL-05-maxsum-direct.md) |\n| `Q-foldL-run-census` | PARTIAL | What ceiling on L does the channel-compatible run census force, and is that ceiling forced or a bridge? | Channel compatibility is exact and cheap but buys little: 3 in 71 at kappa(1) = L, the forced ceiling is a presence-relaxation rather than a multiset-exact bound, exact tiles stop at T_29, and the angle's own verdict sends it to TODO 0c, bounding a residue-deleted maxsum, which this pass did not attempt. | none | [attack-foldL-01-census.md](history/staging/attack-foldL-01-census.md) |\n| `Q-foldL-transport` | ANSWERED | Is there a third invariant of the tile's gap alphabet that provably survives the fold recursion, alongside the sum rules and the maxsum family? | Yes: the tail-count profile transports (the Tail-Count Transport, PROVEN, verified to T29 at 214,708,725 twin slots) and is sharper per fold than the standing route by a factor of three to eight in nats, but it does not chain, dying on the same accumulating index gate-multiplies.md closes. | none | [attack-foldL-03-transport.md](history/staging/attack-foldL-03-transport.md) |\n| `Q-foldL-window5` | MIXED (foldL-window5-prereg.md: OPEN; foldL-window5.md: ANSWERED) | Does the fold extinction law score one decade blind at W = 2e11? | HIT on all four pre-registered criteria (last fold with L >= 2 measured 631 against band [571, 877]; both counts inside their bands), with the hot systematic now confirmed at five of five windows at measured over predicted about 0.82, which by the prereg's own clause belongs in the law as a known overprediction rather than in the acceptance band. | none | [foldL-window5-prereg.md](history/staging/foldL-window5-prereg.md), [foldL-window5.md](history/staging/foldL-window5.md) |\n| `Q-ford-halberstam` | CLOSED | Does Ford and Halberstam's dual decomposition beat Brudern-Fouvry's (2.6), as its authors said it should? | No: the dual is the same inequality, verified to 3.553e-15 at r = 2 and 6.821e-13 at r = 2..7, and their estimation step is better than they claimed, the Rosser boundary layer making their absolute-value bound an identity with loss 0.000000000 at z = 20, 30, 42; the shortfall is the freedom the step gives up, one level per component and no lower sieve. | none | [attack-ford-halberstam.md](history/staging/attack-ford-halberstam.md) |\n| `Q-fractional-retention` | CLOSED | Do Brady and Li's fractional retention rules at kappa = 2 open the adjacent door sift-limit-attack section 7c named? | Do not pursue: the door is real, applies to our system and carries no regress, but half the rule family cannot be applied at kappa = 2 and the applicability window there is empty, both settled from the sources; the unlooked-for find is that Brady's thesis is prior art for our object. | none | [scope-fractional-retention.md](history/staging/scope-fractional-retention.md) |\n| `Q-frontier37` | ANSWERED | What do the fold-L wave's three exact instruments give one level deeper, at fold 37? | The prior-art check moved the target first: the ninth diagonal L(T_31, 37) = 4 and the maxsum_m(T_31) table for m <= 8 were already embedded in the corpus, so this is a re-measurement, not a first; M_alt = G2 is measured at eight folds and proven at none, and the transport ratio's rise 0.8881 to 0.9477 has no mechanism attached. | none | [frontier37.md](history/staging/frontier37.md) |\n| `Q-full-coefficient-average` | PARTIAL | Can aggregating the complete coefficients before a correlation theorem remove the explicit cofactor count, and what additional estimate is needed? | Exact factor and rounded-endpoint Fourier identities retained, with c_(i,0)=3/5 and an explicit composite-filtered weighted sum. The full family is not 1-bounded, but the sufficient phase condition admits at least k=0,+/-1 on the left and l=0,+/-1,...,+/-6 on the right for every Mellin twist; the earlier zero-only claim is corrected. Composite filtering and the required correlation rate remain unmatched. Proposition 6.5, independently reviewed including on 2026-09-09, proves coefficient norm at least (log x)^(2/5) for representations by 1-bounded functions on all smooth inputs. This does not exclude density-one representations, paid growing components or a jointly treated Fourier sum. No sufficient signed twin margin follows. | C | [full-coefficient-average.md](full-coefficient-average.md) |\n| `Q-g2-43-term` | ANSWERED | What is the fourteenth exact term of the G2 ladder, past 41#? | G2(43#) = 618, at least position 830,330,079,152,051 and attained at 8 positions of the period, confirmed by two independent enumerations over disjoint natal masks agreeing on value, multiplicity, position and survivor count; 47# is priced by direct probe at 1.49 days per run and stays out of session reach. | none | [phase1-T2b-exact-ladder.md](history/staging/phase1-T2b-exact-ladder.md) |\n| `Q-g2-falls-decision-rule` | PARTIAL | Does G2(x#)/x^2 fall, and what pre-registered decision rule would settle it? | The instrument is calibrated and sharp, resolving the one-class control at 11.6 sigma on the same nine points where it returns 0.9 sigma on G2, and the honest band on the slope is +/-0.117 at nine terms, so the rule is registered; but the power analysis puts separation at x = 53 only if the falling law is x ln^2 x and at x = 151 if it is x ln^3 x, so phase 1's extra terms settle nothing and no reachable exact ladder does either. | 1d | [phase1-T3prep-decision-rule.md](history/staging/phase1-T3prep-decision-rule.md) |\n| `Q-g2-provenance-x37` | ANSWERED | Does the headline G2 exponent depend on the eight terms this repository has never reproduced, and does any instrument without G2(37#) inside it flag x = 37? | MEASURED, both largely negative: the custody-only refit lands at 1.533 against the headline 1.498 and no provenance-block drop moves the corrected central by more than 0.035 (Ghat(64) alone moves it 0.002), while no G2-free instrument flags 37 in the high direction, the one \\|z\\| past 2 being h(37#) at -2.20 in the low direction and range-dependent; the blind extreme-value forecast of G2(37#) does overshoot at z = +6.58, unexplained. | 0 | [measure-g2-provenance-0829.md](history/staging/measure-g2-provenance-0829.md) |\n| `Q-g2-state` | PARTIAL | What is known about G2, the two-class maximum twin-slot gap, and at what calibration? | The gap is unchanged through roughly fifty recorded passes: the proven upper bound is exponent 4.26645 against a target of 2, and this file consolidates the map rather than computing anything new. | none | [G2-STATE.md](G2-STATE.md) |\n| `Q-gap-spectrum` | ANSWERED | What is the full gap-length distribution of twin slots over one period of the tile, what does the excess functional \\|B_N\\| = sum_{g>N}(g-N) read on an x^s grid, and where does the empirical minimum window count leave zero? | The distribution is exact at nine levels to x = 31 and the pre-registered renewal null FAILS in the light direction at every one: the tail is uniformly steeper than exponential (body slope -1.3302 against -1), no power law survives (the power-law slope moves by a factor 17), the maximum is 2.087 to 2.241 BELOW the renewal maximum, and \\|B_N\\| sits at 0.0245 to 0.7436 of the null and is exactly zero at s = 1.75 where the null predicts 6.5262e+5 empty windows; so no heavy-tail mechanism for a super-linear G2 exponent exists at any computable level, the identity G2 <= N + \\|B_N\\| is proven but degenerates to G2 <= N on this grid, and the sifting onset is flat at 1.63 to 1.70 against a sieve bound f2 that is zero everywhere the grid reaches. | 0 | [gap-spectrum-01.md](history/staging/gap-spectrum-01.md) |\n| `Q-gate-multiplies` | ANSWERED | Does \"the gate multiplies\" close the entire u-frame recursion branch, or only the merge chain? | Only the merge chain and one relative of it: the no-fixed-point argument does not reach TODO 0b at all, though 0b is wrong as stated for an unrelated reason and the error is a factor of ln u; what survives is the copy theorem for maxsum (VERIFIED 40 of 40), which needs a residue-deleted maxsum bound that does not pass through a kill count. | 0b | [gate-multiplies.md](gate-multiplies.md) |\n| `Q-gate-repair` | ANSWERED | Were the verify-the-verifier repairs landed with the gate ending fully green? | COMPLETE: node research/qc.js --full reads FULL GATE PASSED at the close, TOTAL 0 across the 11 checks, selftest 39 known positives firing and 32 controls silent, audit-numbers 246/246, and no bound tail left amber. | none | [gate-repair-finale.md](history/staging/gate-repair-finale.md) |\n| `Q-global-cutoff-averaging` | PARTIAL | Does the wider corpus suggest changing the global decomposition before extending the small-cofactor estimate, and can that change be made without an unpaid transition or outside term? | Derived: independently averaging the two initial Mobius cutoffs over fixed exponent intervals gives S=C2*x+sum G_L(n)G_R(n-2)+O_A(x/log^A x), uniformly on dyadic scales, with all cofactor and prime-power branches included. Each full G has squared norm O(x log x). The existing sharp-corner lower norm then forces cancellation between the corner coefficient and the rest at the same input. This does not estimate their shifted product. The global signed residual was already O(x) by the sieve upper bound and positivity; the new norm representation is not an improved signed bound. Prefer a bounded attempt on this global coefficient pair and a one-sided consumer; the twin margin remains OPEN. | C | [global-cutoff-averaging.md](global-cutoff-averaging.md) |\n| `Q-global-factor-signs` | PARTIAL | Which full factor configurations cause negative global products, can prime-power exceptions be paid, and does a pair-trigger upper bound control the negative part? | Derived: G_i(n)=Lambda_(>W_i)(n)-F_i(s_i(n))*log t_i(n)-E_i(n), with all small-prime factors in s_i and all primes of t_i exceeding W_i. The correction is supported on a small-prime power exceeding W_i; its full shifted effect is O_epsilon(x^(39/40+epsilon)). After also paying proper prime powers, the residual is a composite-filtered signed smooth/rough cofactor sum. A ten-small-prime cell has F=-84 although every pair lies below a, refuting the proposed pair-trigger majorant. Prime filters contribute explicit terms when bounding the negative part; dropping them is an upper bound, not an identity. No scale-x signed improvement or twin margin follows. | C | [global-factor-signs.md](global-factor-signs.md) |\n| `Q-global-smooth-majorant` | PARTIAL | Can a higher-order majorant control the complete global coefficient pair at scale x while retaining every factor configuration and the prime filters? | Derived using the corrected Henriot upper theorem: a C3 probability average of the same admissible initial cutoffs gives a full residual with sum \\|Ghat_L(n) Ghat_R(n-2)\\|=O(x). The three-smallest-prime majorant has bounded harmonic mass and meets the source growth class uniformly, despite not being multiplicative. All prime-power exceptions are paid. This replaces the O(x log x) absolute budget for the earlier logarithmic profile by O(x) for a different admissible profile representing the same signed residual to arbitrary logarithmic precision. The implied constant is not compared with C2 and no improved signed lower bound or twin margin is supplied. | C | [global-smooth-majorant.md](global-smooth-majorant.md) |\n| `Q-greedy-oracle` | ANSWERED | Is the corrected greedy a G2 oracle? | ORACLE ESTABLISHED, 13 of 13 exactly-known terms hit with minimum ratio 1.0000, so the greedy RULE is an exact solver where the answer is checkable at x <= 41; the rider matters more, the search budget needed grows 3.31x per additional prime and two objects whose truth reaches further show the same estimator's fidelity DECAYING, so nothing is established for the greedy AS RUN on the ladder. | 1b (retired) | [greedy-oracle-validation.md](history/staging/greedy-oracle-validation.md), [phase1-T1-greedy-oracle.md](history/staging/phase1-T1-greedy-oracle.md) |\n| `Q-grouped-divisor-moment` | ANSWERED | Does the full gcd-normalized moment proposed by the literature audit hold, and what exact portion of the twin-prime remainder does it control? | The proposed moment is derived from classical completion with all coefficient sectors and uniform twists included. Full rectangles are controlled when delta<19/25 and delta+3nu<161/100, in addition to the preceding region. A concrete extra cut d<=floor(x^(151/200)), de^3<=floor(x^(321/200)) controls the entire d~e~x^(2/5) benchmark and leaves an explicit smaller-domain endpoint remainder. The uniform product threshold stays below 19/25. At delta=8/25,nu=9/20, the next deficit is confined to small-common-divisor nonzero kernels; their required saving and the global twin margin remain OPEN. Finite validation does not prove asymptotic rates. | C | [grouped-divisor-moment.md](grouped-divisor-moment.md) |\n| `Q-growth-law` | ANSWERED | What does G2(x#) actually grow like? | MEASURED about 0.762 * x ln^2 x * lnln x over x = 11..79 (18 terms, ten log-linear families AICc-ranked), with the honest statement being the band and not the winner; the pure power law is excluded against a control that knows its own answer, and G2/h costs one further logarithm. | none | [attack-growth-law.md](history/staging/attack-growth-law.md) |\n| `Q-h2-extension` | CLOSED | Is extending A288815 (h2) past its 21 terms the high-value computation it was taken for? | No, on both halves, and it is recorded in OUTCOMES.md as WITHDRAWN: term 22 is a roughly four-year single-thread computation for the specialists' own tuned code, all three cheaper proxies fail downward and growing (the direction that fakes a smaller exponent), and the control measured one extra term moving a ten-term exponent fit by 0.022. | none | [h2-scoping.md](h2-scoping.md) |\n| `Q-handoff-review-0906` | ANSWERED | Does the handed-back arithmetic campaign survive an independent check of its main regional estimate and the conclusions used to choose the next research direction? | The bounded handoff audit is completed in reports 20 and 21: the checked local reduction and regional mechanisms survive, named source statements were verified, and the consumer, corner support, rate, shrinking-margin and identity-piece overclaims were corrected. Joint Cauchy is now priced and adds no region. The multiplicative band transfer has a separate PARTIAL owner with a weaker continuous-scale payoff. Imported deep theorems remain imports; this is not corpus-wide certification or a twin margin. | C | [handoff-review-0906.md](handoff-review-0906.md) |\n| `Q-hawkins-read` | ANSWERED | Does the Hawkins random-sieve tradition, read at source, hold a two-class variant, a stage-to-stage moment identity, or an answer to the analogue question? | NOT FOUND for a two-class or k-residue variant; the moment identity is found in the null and not in the fold; and the tradition raises the analogue question repeatedly and answers it only as a missing constant. Two primaries stayed paywalled, so the absence claims rest on the rest of the bibliography. | none | [hawkins-read.md](history/staging/hawkins-read.md) |\n| `Q-head-remainder-0830` | PARTIAL | Does the 5.3 percent remainder that Hardy-Littlewood leaves in the head's endpoint deficit Delta = 2 - beta derive by a route that is not HL: the frozen-sieve first-survivor object, a level-of-distribution correction, or the exact tile-ensemble head? | No route that is not HL closes it, and the remainder as recorded is not what it was taken to be. The record's Delta_meas = 0.6214 at [1e7,1e8) pools a decade over which the prime density falls ten percent, and the OLS intercept absorbs the gradient (a Simpson-type effect, sign derived): the two half-decades read 0.6848 and 0.6500, the four quarter-decades 0.7203 to 0.6565, and pooled minus the gap-weighted mean of the halves is -0.0378, -0.0455, -0.0524 at the three decades, so the same-sign 5.3 percent miss is the pooling. At half-decade resolution HL's pipeline prices Delta to 1.2 and 0.4 percent at the top two windows, inside one standard error, and the sign of the miss alternates. That agreement is a near-cancellation of two four-percent effects: the pipeline under-predicts its own rotation ensemble by 3.6 to 4.9 percent (pipe/ens 0.951 to 0.964 across y = 313 to 19997, MEASURED by exact CRT sampling; 1.049 to 0.958 at y = 11 to 23 by full enumeration), an ensemble quantity that derives by finite count with no prime input and whose sign is opposite to hl3 section 2's guess; and the anchored window's deficit sits 2 to 5 percent below its gap-scale-matched ensemble (Delta_meas/Delta_ens 0.878 to 0.995 over six half-decades, 0.954 to 0.995 on the four with more than 13,000 gaps), the anchoring part, which is a three-point prime correlation in a short window and is HL-strength by every route named. The head is a slack field and nothing here touches Z2. | Z4 | [attack-0830-head-remainder.md](history/staging/attack-0830-head-remainder.md) |\n| `Q-head-residual` | PARTIAL | What is the head's residual prime-origin factor h/R = 1.09 -> 1.03, and does it derive? | h - R decomposes exactly and A_forced derives; h/R -> 1 follows from beta CV^2 being bounded, beta CV^2 -> 2 controlling the RATE and not the limit (under it h - R -> D = -0.95, so +8.4661/ln^2 p cannot be the limit law); R is the continuum functional and the integer-origin null sees R + 1/2 exactly, so h - R = 5.679 depends on the unstated null population, reading 5.179 discrete-uniform and 2.925 coprime-to-30; the mod-30 answer is NO by a valid route, class term -0.0001; Delta = 2 - beta is priced by HL to 5.3 percent with nothing fitted and HL's own 1/ln x term absorbs the miss, so the 5.5 s.e. framing is dropped; the asymmetry-as-mechanism claim and the [0.289, 0.361] bracket are refuted, hl3's 16 percent standing; beta_null is 1.898 for the discrete-consistent null and 2 in the continuum; the 72/28 split reads 81/19 in the other exact order; c = 0.844 is 0.722 on a subset, pinned to 15 percent; Delta_HL ln p/lnln p = 4.02 is a property of the prediction; and the N4 plateau is not OPEN, needing beta CV^2 = 2.71 rising to 6.63. | Z4 | [head-residual-factor.md](history/staging/head-residual-factor.md), [head-residual-hl3.md](history/staging/head-residual-hl3.md), [head-residual-null.md](history/staging/head-residual-null.md) |\n| `Q-heath-brown-edges` | ANSWERED | Does the a=1 edge, and hence the corner S_0, come from the particular two-cutoff Vaughan identity in prime-detection-spec.md and shifted-prime-decomposition.md, or does an equivalent edge reappear under Heath-Brown's identity and the other standard decompositions of Lambda? | The exact identities and exponent simplex leave a=1 in separately bounded formal pieces, including Heath-Brown's balanced j=K configuration. This does not prove an intrinsic edge after identity-piece cancellation. The bounded-cofactor object has affine dilations, not literally F(n)F(n-2). Linnik's fixed truncation has a large term-wise absolute tail by an elementary counting proof; its signed tail is not thereby bounded below. Heath-Brown K=3 narrows the classical split band, with coefficient and Type I normalization obligations retained. No full alternative reduction, new region or twin margin follows. | C | [heath-brown-edges.md](heath-brown-edges.md) |\n| `Q-history-dial` | ANSWERED | Does revealing the scour primes' classes concentrate the floor on a bounded set of strike locations, and what is the history dial worth? | Disconfirming: the class axis is linear at 2.90 survivors per prime read, so no bounded set of strike locations carries the floor, and the touch-count ceiling is proven and tight; deliverable (1) of the brief already existed as canonical state (the forcing ladder 16 -> 45 at @11), so nothing new was bought there. | A | [attack-history-dial.md](history/staging/attack-history-dial.md) |\n| `Q-hm-basis` | ANSWERED | Does the u_sup divergence survive the change to the natural (h,m) basis? | It survives: the two bases are nested rather than rivals, the natural one is the outer and weaker member, and the gap between them is a flat factor of 7.26 across nine levels, so the divergence that closed u_sup runs in both bases. | none | [attack-hm-basis.md](history/staging/attack-hm-basis.md) |\n| `Q-holt-2605` | ANSWERED | Does the fifteenth Holt manuscript, arXiv:2605.19165, contain anything that touches our record? | No: it carries no twin, no max-gap bound and no G2, its \\|s\\|/2 threshold restates the 2603.25915 / 1408.6002 inequality, and the interaction check is negative, our extinction law diverging from his by a factor 396 across three decades of W; the sweep is now complete at fifteen of fifteen. | none | [holt-2605-sweep.md](history/staging/holt-2605-sweep.md) |\n| `Q-hsub-reductions` | PARTIAL | Can (H-sub) be proven or refuted, and what does the fold machinery reach? | Neither, but the statement is smaller: three reductions of the hypothesis are proven, the lemma consuming only power pairs so that it weakens to (H-sub-pow) with the conclusion unchanged, the fold machinery's inability to reach any of them is made exact, and the hunt over all 111 reachable pairs found no drifting family and no counterexample. | 1d | [attack-hsub-01.md](history/staging/attack-hsub-01.md) |\n| `Q-hsubpow-K` | CLOSED | Can (H-sub-pow) be proven with an explicit K in the legal zone? | NO by three mechanisms: the CRT lift (K* >= pi(y') - pi(y) diverges), anchored caps (diverge at every base), Iwaniec at two classes (it IS beta2-note); TODO 1d's legal zone is mis-stated, the trusted zone is [1.3946, 11.3568). | 1d | [hsubpow-explicit-K.md](history/staging/hsubpow-explicit-K.md) |\n| `Q-hsubpow-K-0829n` | OPEN | Can (H-sub-pow) be proven with an explicit K inside the trusted legal zone [1.3946, 11.3568) by a mechanism the 2026-08-28 pass did not close? | No K is proven at any base; the single open inequality is the uniform-in-k ratio cap G(b^(k+1))/G(b^k) <= e^K G(b), which is a proof gap at a fixed base and a possible truth gap across bases, since for any law G ~ c n^beta (ln n)^delta the all-bases hypothesis holds with finite K if and only if delta >= 0. | 1d | [attack-0829n-hsubpow-K.md](history/staging/attack-0829n-hsubpow-K.md) |\n| `Q-hybrid-bound` | CLOSED | Does an exactly-handled head of small primes plus a sieved tail improve the certified exponent? | The glue closes and is a real theorem, and it buys one factor of 1.93 in x and nothing asymptotic: the window where the bound beats 4.2665 widens from x <= 227 to x <= 439 and never returns, the head can only reach x0 = O(ln x), and the literal glue the brief named (A144311 as base case) is dead on a type mismatch. | none | [attack-hybrid-bound.md](history/staging/attack-hybrid-bound.md) |\n| `Q-identifications-prior-art` | ANSWERED | Who owns the three identifications that were banked without a prior-art search? | All three are OWNED in their conventions: the complementary-window duality, maxsum_m as the largest m-spacing of a scan statistic, and the alternation language as a textbook charge constraint; the exponent law H = a + b ln D and the H* Shearer-threshold ladder come back NOVEL-SO-FAR on four calibrated channels. | none | [identifications-prior-art.md](history/staging/identifications-prior-art.md) |\n| `Q-import-bfree` | ANSWERED | Does the ergodic theory of B-free systems - Mirsky measures, Toeplitz structure, tautness, heredity, entropy - say anything about the twin comb? | It lands as an exclusion: the comb is exactly the object the field's newest generalisation is built for and that generalisation names the comb's own regime as the one it declines to study, the limit comb is the single point {-1} so every ergodic statement about it is vacuous, and the yield is a corrected address for the walls plus one live external open problem. | none | [import-bfree.md](history/staging/import-bfree.md) |\n| `Q-import-boolean-analysis` | CLOSED | Does the analysis of Boolean functions reach the anchored bias? | Five things that had to hold do not, starting with the product space eating the statement worse than C^pi(x); the forced degree saturates at 2 and the globality hypothesis fails at q1, so the row supplies a wall address rather than a route, and every figure is SCRATCHPAD-GRADE and may not be quoted until re-derived inside an embedded producer. | none | [import-boolean-analysis.md](history/staging/import-boolean-analysis.md) |\n| `Q-import-bridge` | CLOSED | Do Ojaroudi's Bridge and collision lemmas certify a number the corpus lacks? | Both lemmas are true and both proofs go through in three lines here, and the answer to the question the import was raised for is no: the certified-number verdict is zero by an identity rather than by a margin, and the per-prime nature is what stops it. | none | [import-bridge.md](history/staging/import-bridge.md) |\n| `Q-import-bridge-prereg` | SUPERSEDED | Do Ojaroudi's two elementary lemmas, imported onto our tile, beat the incumbent per-prime discrepancy numbers? | Pre-registration only, sealed before any number was computed: the predictions, the incumbents, the win condition and five declared traps are fixed here, and the null result is the expected result because P4 is an identity; the outturn is import-bridge.md. | none | [import-bridge-prereg.md](history/staging/import-bridge-prereg.md) |\n| `Q-import-chaining` | CLOSED | Can Dudley's entropy bound or Talagrand's generic chaining beat the union bound over positions on the maxsum law? | No, and the reason is geometric rather than probabilistic: the entropy integral of the true increment metric is already 1.054-1.099 times rms*sqrt(2 lnW) at its lower branch and 1.484-1.545 at its upper, flat across z = 13..23, so chaining's ceiling with a perfect universal constant sits below the union bound's floor; the honest constant-carrying chain measures 2.878 to 3.027. | 0c | [import-chaining.md](history/staging/import-chaining.md) |\n| `Q-import-distortion` | CLOSED | Does the distortion method of covering systems speak at two classes per modulus, and what does it certify on the window? | It does speak at two classes, and the row's expected blocker was the wrong one: its economy sum 4/p^2 converges (sup 0.364545) where the union bound's sum 2/p dies at x = 13, but it charges in ambient length, its measures living on a CRT product so the certified window is at least a primorial and the only published bridge back to an interval is exponential in the number of progressions. | none | [import-distortion.md](history/staging/import-distortion.md) |\n| `Q-import-distortion-prereg` | ANSWERED | What does the distortion method certify on our window, and does it speak at one class per modulus or at two? | Sealed before any line of the producer existed; scored in import-distortion.md, where the deciding question comes back opposite to the row's expectation: BBMST's criterion carries no distinctness, multiplicity or minimum-modulus hypothesis, so the method speaks at two classes and the one-versus-two-class wall is not its wall. | none | [import-distortion-prereg.md](history/staging/import-distortion-prereg.md) |\n| `Q-import-entropy-decrement` | CLOSED | Does Tao's entropy decrement argument, the one place a parity-type obstruction was circumvented for a two-point correlation, transfer to the tile's census? | It does not: the engine runs on multiplicativity at small primes and the census is defined by the absence of small prime factors, Tao scopes twin-prime sums out himself in print, and the census sits on the structured side of the pretentious dichotomy; all figures SCRATCHPAD-GRADE. | none | [import-entropy-decrement.md](history/staging/import-entropy-decrement.md) |\n| `Q-import-fracparts` | ANSWERED | Does Saffari-Vaughan Theorem 10 cover the branches carrying the Skeleton door's mass? | Prediction HIT: the covered branches carry at most 8 percent of the mass; anchor and a sharper wall-address banked; the THEOREM column is struck, not merely overpriced, because Theorem 10's range condition is x^{6/11+eps} < y, which fails at fixed M; no route; no cap-36 line is upgraded. | 4 (retired) | [import-fracparts.md](history/staging/import-fracparts.md) |\n| `Q-import-hypergraph` | MIXED (import-hypergraph-prereg.md: OPEN; import-hypergraph.md: ANSWERED) | Does hypergraph covering (Pippenger-Spencer / FGKMT) apply to the adversary's construction? | The covering theorem is class-count-agnostic and the analogy is EXACT at the hypothesis level, leaking at exactly one named place, the edges the machine needs as input; Branch A landed a theorem-provenance lower bound G2(x#) >> x ln x (HELD), Branch B died as predicted with a sharper wall than the registered one, and none of the three pre-registered kill lines fired. | none | [import-hypergraph-prereg.md](history/staging/import-hypergraph-prereg.md), [import-hypergraph.md](history/staging/import-hypergraph.md) |\n| `Q-import-interp` | ANSWERED | Does the Bayati-Gamarnik-Tetali interpolation method reach the G2 Fekete object? | No: BGT closes on hypothesis H1, since the constraint count never splits (pi(st) - pi(s) - pi(t) is zero at no pair with st >= 25, minimum 2, maximum 12, VERIFIED on 104 pairs), and the one coordinate that does split has limit +infinity; the row lands as WALL-ADDRESS plus the identity that the submultiplicativity defect is the Overshoot slack. | 1d | [import-interp.md](history/staging/import-interp.md) |\n| `Q-import-interp-prereg` | SUPERSEDED | Does the interpolation method (Bayati-Gamarnik-Tetali, Proposition 5) transfer to the G2 ladder, and in which coordinate? | Pre-registration only, committed alone before any producer existed, with the source verified at page images first (Proposition 5 is a special case of de Bruijn-Erdos, by the authors' own attribution); the predictions and the meaning of each outcome for the map are fixed here, and the outturn is import-interp.md. | none | [import-interp-prereg.md](history/staging/import-interp-prereg.md) |\n| `Q-import-kw-zonegap` | ANSWERED | What does the graded Kourbatov-Wolf row buy the Z2 programme, and does the flagged 0.7574 against 0.7504 agreement corroborate anything? | A split-cell circularity verdict: the trend formula's raw performance is real, G/Tbar inside [0.78, 1.05] over thirteen decades with zero fitted parameters, but the flagged agreement is adjudicated a range accident, and since Theorem occurs zero times in the source the row can only be priced PUBLISHED-ANCHOR plus WALL-ADDRESS, whose wall is a quantifier: an almost-all over a log-sparse record sequence cannot deliver an every-zone statement. | none | [import-kw-zonegap.md](history/staging/import-kw-zonegap.md) |\n| `Q-import-l1l2` | CLOSED | Does an l1 -> l2 conversion on the (e,a) coefficient sum buy the needed sqrt(log)? | The pre-registered kill line (rho >= 0.5 at every level) does NOT fire, rho falling 0.4352 to 0.0645 at z = 13..41, and the programme dies anyway on what the decay measures: \\|\\|Theta\\|\\|_1/\\|\\|Theta\\|\\|_2 grows by 2.01 per added prime against 2.0516 for S_sat, so the l1/l2 gap is not a proxy for the C^{pi(z)} price, it IS that price; rows 5 and 6 retire together. | none | [import-l1l2.md](history/staging/import-l1l2.md) |\n| `Q-import-l1l2-prereg` | CLOSED | Does the l1 to l2 conversion on Theta_e(a) reach the threshold the import map set? | Sealed before any number was computed; scored in import-l1l2.md, where the pre-registered kill does not fire and the programme dies anyway: rho falls monotonically from 0.4352 to 0.0645 over z = 13..41, never reaching the map's 0.5 line, and what the conversion outputs is the sharp maximal law. | none | [import-l1l2-prereg.md](history/staging/import-l1l2-prereg.md) |\n| `Q-import-map` | PARTIAL | Which foreign mathematics is worth aiming at this object, and what did each landed import actually buy? | Nineteen landings, zero routes opened on the exponent; every one banked a proven identity, a derived constant, a published anchor or a wall address, which is why payoff is graded by TYPE rather than by probability, and the map is live with rows still unrun. | none | [IMPORT-MAP.md](IMPORT-MAP.md) |\n| `Q-import-map-construction` | ANSWERED | What was searched, what was rejected, and at what verification level, in building IMPORT-MAP.md? | The reconnaissance record behind the map: candidates rejected with reasons (hypergraph containers for want of a supersaturation input; determinantal processes by a proof-level obstruction, the twin tile failing the pair inequality at d = 6), a per-attribution SOURCED / SOURCED-BIB / MEMORY ledger, and two results derived here and offered for adjudication. | none | [import-map-construction.md](history/staging/import-map-construction.md) |\n| `Q-import-maxplus` | ANSWERED | Do max-plus (tropical) spectral theory and subadditive ergodic theory bound the fold recursion's growth exponent? | The mapping is exact rather than an analogy - the fold is p-fold cyclic duplication followed by max-plus state elimination and the copy theorem is a semiring identity, verified at 6 of 6 folds - but the tropical Perron root of the tile is the mean gap mbar, not G2, so the import delivers a structural negative, a sharpening of A5's Theorem A, and refuses the growth exponent. | none | [import-maxplus.md](history/staging/import-maxplus.md) |\n| `Q-import-proof-complexity` | CLOSED | Does proof complexity supply the barrier that wall-note Face 4 wants? | The row is VOCABULARY-ONLY and is recommended for rejection at the gate: the shared word is parity naming two different objects, a degree bound is a pseudo-expectation whose finite exact instance the corpus already refuted, and any unconditional barrier would imply Siegel-zero finiteness; the payoff is Tao and Brady 2014 read at source as a published anchor and a wall address, not a theorem. | none | [import-proof-complexity.md](history/staging/import-proof-complexity.md) |\n| `Q-import-repulsive` | CLOSED | Does any repulsive point-process model (determinantal, Coulomb gas, hyperuniform) reproduce the twin tile's sub-Poisson counts? | NO, two-sided: g(2) = 0 kills positively-associated models and g(6) = 2.66 (g never in (0,1); support 6 \\| d, g >= 2.38) kills Hermitian determinantal and negatively-associated ones; the Gibbs extension is withdrawn; every fixed level is periodic hence class I hyperuniform and the statement is empty, while along the diagonal family L = y^u no member approaches a hyperuniform limit in Var/E, which is a statement about the family and not a class for any one tile; the determinantal row's quoted product evaluates to 0 and its obstruction distance is 6 not 2. | none | [import-repulsive.md](history/staging/import-repulsive.md) |\n| `Q-import-rough-anatomy` | ANSWERED | Does the analytic theory of rough numbers derive the depth law's drift, and is u = 2 a singular point of the dimension-2 sifting functions? | SCRATCHPAD-GRADE throughout, and leaning on three never-red-teamed premises: the drift is derived to 0.16 per cent by exact Mertens products and NOT by any published next-order term (the one in print is the wrong term for this object), and u = 2 is one identity seen four times rather than a singular point of the dimension-2 sifting functions. | none | [import-rough-anatomy.md](history/staging/import-rough-anatomy.md) |\n| `Q-import-rows-15-17` | ANSWERED | Which three new fields deserve import-map rows, and at what price? | Row 15 is a wall address and a published anchor and NOT a theorem, since Theorem 10 requires x^{6/11+eps} < y and fails at every fixed M; row 16 is stale, its experiment already executed by record-location-null.md and its prior-art half closed by lit-kourbatov-shortfall.md; row 17 lands as priced; six fields rejected with pointers and one promoted to row 17. | none | [import-map-rows-15-17.md](history/staging/import-map-rows-15-17.md) |\n| `Q-import-scanstat` | ANSWERED | Does the extreme-value theory of scan statistics settle the disputed sqrt(m) claim of IMPORT-MAP row 1? | The identity offered for adjudication is true and its conclusion is false - Sum gamma(k) = 0 is trivial while the step to Var(S_m)/m tending to 0 is refuted by a free permutation counterexample - the Leadbetter half of the row is confirmed as written and the pre-registered kill criterion passes at both target levels, one of them blind; the adversary pass corrected the claim that T29 had never been computed here to a reproduction. | none | [import-scanstat.md](history/staging/import-scanstat.md) |\n| `Q-import-scanstat-prereg` | OPEN | What is the maxsum exponent at T_23 and T_29, predicted from a T_13/T_17/T_19 fit before the run? | Machine-written pre-registration only: the extrapolation laws, the grid and the kill criterion (Model A must beat Model B on the ln-RMS of excess_m at T_23) are fixed before any scoring, so part 4 cannot be scored against a prediction chosen after the fact. | none | [import-scanstat-prereg.md](history/staging/import-scanstat-prereg.md) |\n| `Q-import-shearer` | ANSWERED | Where does Shearer's region actually end for the fold read as a repulsive lattice gas? | SURVIVES WITH CORRECTIONS under the 2026-08-19 adversarial pass: the complete-graph identity is verified by independent-set enumeration at n = 2..8, H* = 35..481 reproduce exactly and the 2/sqrt(e) floor is sound, but the tightness witness's disjoint-intervals sentence is false exactly where invoked and the atomicity lemma's hypothesis is load-bearing and not shown, the closure surviving instead on the sequel's gamma_i >= mu(f_i). | none | [import-shearer.md](history/staging/import-shearer.md) |\n| `Q-import-shearer-prereg` | ANSWERED | Where does Shearer's region actually end on the repulsive lattice gas model of our window? | Sealed before any line of the producer existed; scored in import-shearer.md, where the exact criterion says the local lemma is the union bound on a complete dependency graph, and the two levels it buys over the local lemma's own sufficient condition are the only thing it buys. | none | [import-shearer-prereg.md](history/staging/import-shearer-prereg.md) |\n| `Q-import-sofic` | ANSWERED | Does constrained coding and symbolic dynamics identify the Alternation Lemma's shift, and does the import map's shape claim survive? | The identification is exact and better than the map states - the shift is strictly sofic, two states, four edges, Perron root exactly 2, and the graph's own per-step rate is 2/p rather than the 3/p the map divides by - and the pre-registered ratio test then kills the map's literal formula while the graph-corrected one survives only as a small-D transient spent by fold 37; the shape claim does not stand, and it fails in the corpus's favour. | none | [import-sofic.md](history/staging/import-sofic.md) |\n| `Q-import-sofic-prereg` | OPEN | Does the constrained-coding presentation of the fold predict the exact per-fold L, and is the shift sofic? | Pre-registration only, and declared weaker than the G2(41#) case because the nine exact L values were already published in the corpus: what is fixed here is the estimator, the exclusion rule, the trend statistic and the kill thresholds, plus a soficity verdict written before the check, strictly sofic and not of finite type. | none | [import-sofic-prereg.md](history/staging/import-sofic-prereg.md) |\n| `Q-import-stein` | CLOSED | Does Stein's method for Poisson approximation derive the extinction law's constants and the ~3.8 joint deficit? | The pre-registered kill fires at both targets for one structural reason: each object is driven by exactly one shared uniform draw, so the non-neighbourhood indicators reconstruct it and b3 climbs to its own ceiling, measured at 0.850 to 0.9998 of that ceiling and 200 to 800 times b1 + b2; with no error term there is no Stein derivation, and the first-moment arithmetic banks the (A, c) pair anyway, relabelled. | none | [import-stein.md](history/staging/import-stein.md) |\n| `Q-import-stein-prereg` | CLOSED | Does Stein's method for Poisson approximation bound the extinction law or the ~3.8 constant? | Sealed before any producer ran; scored in import-stein.md, where the pre-registered kill criterion fires at both targets for one structural reason, that both objects are driven by exactly one shared uniform draw, which is precisely the dependence the b3 term measures. | none | [import-stein-prereg.md](history/staging/import-stein-prereg.md) |\n| `Q-import-suen` | ANSWERED | Do Suen's inequality, Janson's inequality and the Lovász Local Lemma family supply the missing proof for the anchored delta? | No inequality can: the import lands and what it certifies is that the target was never a lemma, the NOT-REACHED item being the Twin Prime Conjecture wearing a percentage; what it does deliver is the constant in the sharp form and the wall's address in local-lemma coordinates. | none | [import-suen.md](history/staging/import-suen.md) |\n| `Q-import-thinning` | ANSWERED | Do point-process thinning and interval coalescence give H'' a null model? | Yes and exactly: the fold recursion is an exactly solvable thinning whose gap pgf transforms by a Mobius map fixing 0 and 1, so under independent thinning the gap word of T_x is exactly geometric with mean mbar/6 at every rung with no error term, which turns both of the record's fitted parameters into derived ones. | 0b | [import-thinning.md](history/staging/import-thinning.md) |\n| `Q-import-transference` | CLOSED | Does the transference principle and relative Szemeredi theory reach the twin pattern? | VOCABULARY-ONLY and closed by an exclusion in print: the founding paper names our configuration as out of scope in one printed sentence, the theorem degenerates at k = 2 which is the case we want, and the W-trick removes the tile before the argument starts; re-entry costs a finite-complexity system whose solvability implies the Zone Postulate. | none | [import-transference.md](history/staging/import-transference.md) |\n| `Q-import-vc-nets` | CLOSED | Does VC dimension, epsilon-net or set-cover duality bound G2? | VOCABULARY-ONLY, at a self-set 0.2 per cent confidence of opening the 4.2665 -> 2 gap and with every number SCRATCHPAD-GRADE: the range space's VC dimension is small and its growth is O(log log L) (proven), but the dual shatter function is exponential, the interval formulation's hypothesis contains its conclusion, and the full-period formulation is clean, free and on the wrong side of the inequality. | none | [import-vc-nets.md](history/staging/import-vc-nets.md) |\n| `Q-ioslack-survey` | CLOSED | Does any instrument in the corpus carry a sharp-level signature that would let the infinitely-often slack be spent? | None does: 23 instruments surveyed and 19 eligible, smallest per-instrument p_FWE = 0.1082 and 0.8865 after Šidák, so dial 4's fourth channel closes and the dial is finished; the second finding is that maxsum_{L+1} is sharp exactly when L = 1. | 0e (retired) | [ioslack-survey.md](history/staging/ioslack-survey.md) |\n| `Q-joint-correction-source-audit` | ANSWERED | Does completing the supported prime-partner correction yield cancellation beyond the existing Vaughan reduction, and which literature inputs actually justify that completion? | CHECKED prior-art matches for subset convolution, complementary sieve sums and signed smooth-number distribution. DERIVED reconstruction identifies the unrestricted factor weight with an existing short divisor approximant, up to the paid prime-power error. Its supports are x^(12/25) and x^(1/10). Reassembling the entire joint correction returns the already-open Vaughan remainder and supplies no new signed bound. A formal factor-vector control distinguishes the unrestricted logarithmic weight from the selected prime-cofactor branch. The matched general sieve theorems require inputs not supplied here. Literature novelty is unestablished; twin infinitude remains OPEN. | C | [joint-correction-source-audit.md](joint-correction-source-audit.md) |\n| `Q-joint-factor-estimate` | ANSWERED | Can an additional estimate for actual factor configurations control the full signed balance of the C3 residual? | DERIVED: for the single-large-prime delta-rough family F, the residual splits exactly as R_F = C2*x*H_delta - Z_F^p + o(x), where the Type I part C2*x*H_delta is evaluated by ordinary prime BV on each cofactor interval and the twisted Mobius constant 2C2 (corrected 2026-09-08 after V2: Wu Lemma 2.3 is squarefree-only and did not cover the non-squarefree b_R moduli; prime BV covers all moduli), and Z_F^p is the signed prime-partner sum. The complement carries the opposite Type I part -C2*x*H_delta, so grouping by factor configuration transfers Type I mass and leaves the twisted prime correlation B_L unchanged. With the matched upper sieve the family bound is C2*x(1+H_delta-(40/9)H^+), certified below -2/5 C2*x: INSUFFICIENT BOUND. Even a sieve constant 2 in place of 40/9 needs H^abs<1 and leaves the complement unpaid. Independently reviewed 2026-09-09 with the uniform delta-range and complement corrections below. Twin infinitude and every sufficient margin remain OPEN. | C | [joint-factor-estimate.md](joint-factor-estimate.md) |\n| `Q-k4direct-swap` | ANSWERED | Does swapping the shipped Monte Carlo estimator for k4direct meet the @13 gate, and is the x3600 on record a speedup? | The swap meets the gate by x3632, but the margin on record is an ACCURACY margin and not a speedup: k4direct is a 4x to 8x SLOWDOWN, 93.5 s against 11.7 to 23.4 s, and the 198.9 s on record for it is 2.1x too expensive, which drags the @17 extrapolation from 8.5 days on ten cores to about 4.4. | none | [verify-k4direct-swap.md](history/staging/verify-k4direct-swap.md) |\n| `Q-kappa-not-L` | CLOSED | Can L, or its generalisation kappa(m), be bounded well enough to be the route's target? | L is not the target: three bounds on L are PROVEN through the Alternation Lemma and transfer to kappa(m), but the wall is located precisely and the lower bound never becomes exact, so L cannot be made irrelevant; closed 2026-08-19, L has no law of its own, being G2 divided by the block's own spacing and super-additive on disjoint sets. | none | [kappa-not-L.md](kappa-not-L.md) |\n| `Q-kernel-sign-control` | ANSWERED | At delta=8/25, nu=9/20, is the moment Mfrak and its small-common-divisor kernel X_small measurably smaller with the actual coefficient b_u=A_right(gu) than with \\|b_u\\| or with \\|b_u\\| times an independent seeded random sign, and does the actual/random ratio decrease with x over the feasible dyadic range? | MEASURED, negative. Over seven dyadic scales j=18..30 and four boxes, the pre-registered slope of log2(actual/random) for X_small has mean + sd >= 0 in all four families, so the run gives no heuristic support for extra Mobius-sign cancellation in the small-j kernel. The actual coefficient's \\|X_small\\| ranks like one more random draw. The measurement is weak where it matters: the finite moment is 97.66-102.58 percent equal-frequency class, X_small carries 0.02-2.88 percent, J0=floor(x^(1/20)) is 1 at j=18 and 2 at j=20..30, and Z=2 collapses the prime-power sector of (13) to r=2. Equation (21) and the global twin margin remain OPEN and untouched. | C | [kernel-sign-control.md](kernel-sign-control.md) |\n| `Q-kk-substitution` | ANSWERED | Does the Kalmynin-Konyagin construction survive substituting the two-class set Omega_p = {a_p, a_p - 2}, and what lower bound does it give? | STANDS WITH CORRECTIONS: G2(P(y)) >> y (ln y)^3 (lnlnln y)^2 / (lnln y)^4 for y >= y0 survives every attack made here, including an exhaustive brute force of the Proposition itself over four million values of i; five corrections follow and none touches the exponent. | none | [attack-kk-substitution.md](history/staging/attack-kk-substitution.md), [verify-kk-substitution.md](history/staging/verify-kk-substitution.md) |\n| `Q-klz-forward-walk` | ANSWERED | Does the forward citation graph of arXiv:2205.08273 hold a two-class or k-class complexity result, and is Part II out? | The forward graph has exactly one member on three independent indices and it never treats word complexity; no citing work goes beyond one class per prime, Part II is NOT OUT as of 2026-08-19, the owning convention was OEIS A023192 all along, and the comparison is APPLES-TO-APPLES so the exponent-shape sentence needs no heredity caveat. | none | [klz-forward-walk.md](history/staging/klz-forward-walk.md) |\n| `Q-kstar-drift` | ANSWERED | Is the drift in K* across doubling steps structural, and how far past the scannable levels does the certificate reach? | Structural and it lands early: K* is now enumerated at 16 doubling steps against 11, the raw drift slope steepens from 0.6881 +/- 0.1328 to 0.8184 +/- 0.0908, and the period-free certificate touches K*+1 = 18, 93.5% of 2^beta2 = 19.2455, at the base-2 chain step s = 16 itself. | none | [attack-kstar-01.md](history/staging/attack-kstar-01.md) |\n| `Q-kstar-prereg` | OPEN | What is K* at the three next doubling steps, predicted before any period walk? | Pre-registration only, committed alone: the predictions, the scoring rule and the growth-type verdict thresholds are fixed in advance, with the inclusion-exclusion engine validated against an independent scan engine on all eleven known steps first. | D | [attack-kstar-01-prereg.md](history/staging/attack-kstar-01-prereg.md) |\n| `Q-l1-residue` | CLOSED | Is the L = 1 residue count an open counting hypothesis whose proof would deliver the Zone Postulate? | REFUTED as a hypothesis: restore the dropped L >= 2 terms and it IS the Zone Postulate, so the chain is a tautology, and leave them out and what remains is provably no stronger than the target; Lemma A (inside B(p) the residue condition is the kill condition) is PROVEN and VERIFIED at 237 folds, and the route lands back on the residue-deleted maxsum without advancing it. | none | [attack-l1-residue.md](history/staging/attack-l1-residue.md) |\n| `Q-lambda-ledger` | CLOSED | Does parity information survive in the lambda-weighted fold ledger? | REJECT, against a pre-registered threshold fixed before the producer existed: the lambda multiplier exists and is parity-free, so the ledger is blind at its design point rather than at its margins; two columns are pinned by proof, and nothing here is novel mathematics. | Z5b (retired) | [attack-lambda-ledger.md](history/staging/attack-lambda-ledger.md) |\n| `Q-ledger-extraction` | ANSWERED | Can the QC engine's three suppression ledgers be moved out of the checks and into data? | Moved: all three now live in research/qc/ledgers.js, the fast gate and the selftest are byte-identical to their pre-change output, and all 25 entries were verified verbatim against git HEAD by key, order and reason text. | none | [audit-ledger-extraction.md](history/staging/audit-ledger-extraction.md) |\n| `Q-ledger-forced` | ANSWERED | Which properties of the fold ledger's columns are FORCED rather than expected? | Fourteen forced constraints, all identities or wrong-direction bounds; by_new is forced only in the prime gap (its O(1) is MEASURED); nothing forces net from below. | Z5b (retired) | [fold-ledger-forced.md](history/staging/fold-ledger-forced.md) |\n| `Q-left-divisor-signs` | PARTIAL | Can the Möbius structure of the left coefficient A_left(gm) replace the first Cauchy inequality, so that its sqrt(M) loss is not paid, and what does that buy at the target box (delta,nu)=(8/25,9/20) and at the corner a=b=1? | Partly, and not where it is needed. Two lemmas are derived: a Type I bound with no zero-frequency budget, usable at the target box whenever the arbitrary block is shorter than x^(1/4), and a Type II bound that is uniform in the Perron twist heights. Applied to the actual coefficient they confine the target-box deficit to sectors with left prime power r>=x^(43/200) and right prime power q>x^(3/100), and lower the worst block exponent there from 103/100 to 41/40. The corner a=b=1 is untouched: grouped, Type I, Type II, Bettin-Chandee/Wright and even a hypothetical per-block square-root bound all exceed 1 there. No region is added, E_dagger and the consumer (20) are unchanged, and the global twin margin remains OPEN. | C | [left-divisor-signs.md](left-divisor-signs.md) |\n| `Q-lemmaV-meansquare` | PARTIAL | Does the mean-square-in-x form of Lemma V hold, and what does the proof cost? | Proved unconditionally, _H <= B(z,s)*H for every H, z and level s, and the proof stops at exactly one place, B(z,s) bounded: B(z,3.0) measures 1.3833 to 1.4883 at z = 13..37, flat to 8 per cent over the whole computable range, which is a measurement and not a bound. | none | [attack-beta2-01-lemmaV-meansquare.md](history/staging/attack-beta2-01-lemmaV-meansquare.md) |\n| `Q-lemmaV-neighbours` | ANSWERED | What do the five Lemma-V-adjacent instruments give when read at source, and which hypothesis do we violate? | The right instrument is Maynard's Lemma 6.12, owned by Deshouillers-Iwaniec 1982 Theorem 12 and sharpened by Pascadi's Corollary 18 in exactly the term that binds us; the hypothesis violated is narrower and sharper than the corpus expected, since all three papers allow arbitrary rough coefficients in three of five variables and what they require is smoothness where we have none. | none | [lemmaV-neighbours.md](history/staging/lemmaV-neighbours.md) |\n| `Q-lemmaV-usup` | ANSWERED | Does u_sup plateau below beta2, or diverge? | It rises at every added level, 2.0617 at z = 13 to 3.2026 at z = 43: a plateau below beta2 is rejected and so is the fast divergence attack 1 feared, leaving \"plateau above beta2, or divergence\", which for this route is the same verdict (divergence since REFUTED at fixed s by attack-0829n-rml-proof.md section 4.1, adjudicated in verify-0830-usup-convention.md). | none | [lemmaV-sup-extension.md](history/staging/lemmaV-sup-extension.md) |\n| `Q-level-ledger-tight` | PARTIAL | How much of the Level Ledger's 2*3^{pi(x)-1} class-discrepancy bound can be taken back? | The sup-versus-variance gap is not closed, and section 4 shows no L1 Fourier method can close it, since such a bound is already >> 2^{pi(x)} from its own j = 1 term; what replaces the exponent is an exact reduction to one finite quantity, computed by exhaustion to y = 19, and a proven CONSTANT factor 81.0 uniform in x, p and a (53.9 from the cheap y = 17 seed), which does not grow. | none | [level-ledger-tight.md](level-ledger-tight.md) |\n| `Q-lichtman-decomp` | ANSWERED | What is the wall's published constant 3.29956 made of, and is there technique slack a method of this class could spend? | The split is multiplicative, 3.2995525 = 2 x 1.649776 with the second factor an equidistribution deficit, so the brief's \"remaining 1.65 of technique\" has no referent and category (b) is empty; the floor for this class is 2 on Type-I data and still 1.122919 at unlimited level, all arithmetic SCRATCHPAD-GRADE. | none | [attack-lichtman-decomp.md](history/staging/attack-lichtman-decomp.md) |\n| `Q-lit-evans` | ANSWERED | Does Evans, Correlations of almost primes, reach y about x^0.28, does it ever fix h, and is there a secondary term to harvest for the depth law? | None of the three, and the item named the single highest-value unread item in the survey yields nothing for the depth law: over the whole 35-page extraction Buchstab, rough, sifting and omega(u) each occur zero times, so the paper has no sieve-depth coordinate, and its object is E2 numbers, a multiplicative condition rather than a roughness one. | none | [lit-evans.md](history/staging/lit-evans.md) |\n| `Q-lit-fgkt-maier` | ANSWERED | Do the corpus's FGKT and Maier quotations hold against the PDFs of record? | The quotations are VERBATIM where checked, including at the sub/superscript level HTML destroys, and the defects are in the claims built on them: a Buchstab dimension error at four sites, one of them publication-track and self-contradictory, and two provenance errors (the Maier-Pomerance equality form and the FGKT/FGKMT R-statement); Cheer-Goldston was not obtained, so one quotation stays unverified. | none | [lit-pdf-fgkt-maier.md](history/staging/lit-pdf-fgkt-maier.md) |\n| `Q-lit-fkmpt-quotes` | ANSWERED | Do the corpus's FKMPT quotations hold against the journal PDFs of record? | All three quotations are VERBATIM or CONFIRMED and two of the three surrounding claims are supported; the third is not, the cited section 1.3 does not exist (the passage is 1.1) and the connective overreaches, and five corrections are drafted including one silent elision inside a block labelled verbatim. | none | [lit-pdf-fkmpt.md](history/staging/lit-pdf-fkmpt.md) |\n| `Q-lit-halberstam-richert` | ANSWERED | Do the Halberstam-Richert and DHR citations in dhr-verification.md and two-class-lower-bounds.md check out against primary sources? | Mostly reachable, against the brief's expectation: seven claims verbatim against source including beta2 about 4.266 at the book page itself, one UNREACHABLE (HR Corollary 2.4.1), the meaning of DHR and the Friedlander attribution both settled, and 4 defects among 21 further attributions swept. | none | [lit-pdf-halberstam-richert.md](history/staging/lit-pdf-halberstam-richert.md) |\n| `Q-lit-kalmynin-konyagin` | ANSWERED | Do the corpus's Kalmynin-Konyagin citations hold against the PDF of record? | Item 2 is clean, reproducing their Theorem 1 at l_f = 2, h_f = 0, M(f) = 2 exactly; item 1 is MISROUTED, the quote being FKMPT's Definition 1 rather than K-K; and the word twin appears zero times in every version, with the calibration running opposite to expectation, since they delete two or more residue classes per prime and run a dimension-2 Mertens ledger, which our own absence rows still call ABSENT. | none | [lit-pdf-kalmynin-konyagin.md](history/staging/lit-pdf-kalmynin-konyagin.md) |\n| `Q-lit-kourbatov-grob` | ANSWERED | Do the Kourbatov and Grob quotes on which the ZONE-POSTULATE retraction rested hold against the PDFs of record? | They hold: four verification items land VERBATIM against the published article and Grob's three negatives are confirmed, so the retraction was justified and a true claim was not retracted on false evidence; the corpus under-reports Grob in two places. | none | [lit-pdf-kourbatov-grob.md](history/staging/lit-pdf-kourbatov-grob.md) |\n| `Q-lit-pdf-bft` | ANSWERED | Do our Banks-Ford-Tao quotations survive checking against the PDFs of record? | All three quotations VERBATIM, checked by scripted normalised diff rather than by eye; one was misattributed by the brief and not by the document (it is Maier-Pomerance), the demotion resting on their model R stands, and the corrections owed to the record are listed. | none | [lit-pdf-banks-ford-tao.md](history/staging/lit-pdf-banks-ford-tao.md) |\n| `Q-lit-pdf-holt-rudd` | ANSWERED | Do our Holt and Rudd citations survive checking against the arXiv PDFs of record? | Every quotation VERBATIM with no misquotation found, but Lemma 3.1 is not the lemma the demotion needs, the 2p spacing result carries no numbered theorem of that name, several attribution defects and two internal conflicts are ours to settle, and the absence claim remains an absence claim. | none | [lit-pdf-holt-rudd.md](history/staging/lit-pdf-holt-rudd.md) |\n| `Q-lit-scourfield-2008` | ANSWERED | Does Scourfield, Smooth divisors of polynomials (LMS Lect. Note Ser. 352, CUP 2008, 286-311), supply the divisor-distribution estimate that varE-theta2-proof.md section 6 leaves open? | DOES NOT DELIVER, at SECOND-HAND rung, and the chapter is still UNREAD at the page: ten channels were attempted, nine answered, and every one that answered is closed, degraded or paywalled, but the author's own restatement of the 2008 theorem, read at the page image of Scourfield, Funct. Approx. 55.1 (2016) 84 eq. (1.1), counts divisors m <= x of f(n) for n <= x with weight 1, for f a product of pairwise coprime irreducible factors each of degree at least 2, and the open step needs divisors n > 2L above the window, carrying a branch-dependent weight of total mass 4^omega, for the product of three LINEAR factors C(C-2)(C+2). Two gaps each independently fatal, plus a third that is AMBIGUOUS between two author-sourced statements and does not carry the verdict. One correction to lit-smooth-divisors.md is earned: the precision register of the owning convention is not uniformly (log y)^{o(1)} or order-of-magnitude, since (1.1) is an asymptotic with relative error O(1/log x), which is the right register and one epsilon short of the o(1/ln y) the open step needs. Step 1 of Var/E does NOT become conditional on a citation, and no absence may be written, since the chapter's interior lemmas were not read. | 9 | [lit-scourfield-2008.md](history/staging/lit-scourfield-2008.md) |\n| `Q-lit-siebert` | ANSWERED | Is Siebert's 1983 result stated for general sieve dimension kappa, or for kappa = 1 only, and does it therefore already contain the kappa = 2 extremal example that SEARCH-CONVENTIONS section 3 records as not found? | kappa = 1 ONLY, on three independent statements of the theorem (the chapter's own publisher abstract, Granville 2020 section 1, Friedlander-Iwaniec 2022 page 20), all naming the LINEAR sieve and the Jurkat-Richert f and F; the chapter's full text was NOT reached, so this is testimony not primary reading, and the SEARCH-CONVENTIONS section 3 \"no published kappa = 2 extremal example\" row needs no correction. | 0 | [lit-siebert-1983.md](history/staging/lit-siebert-1983.md) |\n| `Q-lit-wu2004` | ANSWERED | What is H_theta(2) in Wu 2004, and what does the source say about theta approaching 1? | The source does not answer it: Wu never parameterises H by a level of distribution at all, so the theta to 1 reading is not in the paper; three defects follow in what attack-lichtman-decomp.md quotes, one of them a wrong inference built on a wrong table row, and the only proven cap is H <= 1. | none | [lit-wu2004.md](history/staging/lit-wu2004.md) |\n| `Q-localized-gap` | CLOSED | Does the Localized Merge Lemma close a chain from the localized twin-slot gap to a bound on the maximal gap? | The chain does not close, and it fails on a one-line averaging argument rather than on any sieve question; what survives is the growth law of maxsum_m and the boundary question, closed at (a)/(b) = (a)/(c) = 1.0000 at every x and every m <= 1024. | none | [LOCALIZED-GAP.md](LOCALIZED-GAP.md) |\n| `Q-lp-push-x43` | ANSWERED | Does the level-D LP's DP1 > DP3 inversion survive being pushed to x = 43? | Yes, and the correction strengthens it: the two statistics still climbing at x = 23 stop climbing, the pooled floor moves UP from 3.1945 (16 readings) to 3.3152 (33 readings), DP1 reads 2.32 of the 3.2665 against DP2+DP3's 0.95, and every one of the 33 pooled readings clears the (1 + beta_2)/2 = 2.6332 the inversion needs, the tightest at 2.6692. | none | [lp-push-x43.md](history/staging/lp-push-x43.md) |\n| `Q-maier-matrix` | CLOSED | Can the Maier matrix transfer exact residue-level knowledge about twin slots into an interval statement, and if not, where does it stop? | It stops at a scale collision: at q = x# the matrix is the identity 0 = 0, at q = y# with y < x the transfer identity is PROVEN and VERIFIED 20 of 20 but is a first-moment identity only, almost-all would need the second moment which is the localized twin-gap problem itself, and Granville-Soundararajan Corollary 1.4 is vacuous at every computable scale. | none | [maier-matrix.md](maier-matrix.md) |\n| `Q-maxgap-law` | ANSWERED | Does maxgap ~ c*mbar*lnD survive far enough to contradict Maier-Pomerance, and why did two agents measure two constants? | It does not, and not because the law breaks late: it was never an extreme-value law but a curve through a two-dimensional surface, and the two agents sat on different curves without recording which; on the diagonal c really is flat over a lever 46 times long with an estimator demonstrably able to see Maier-Pomerance growth, and one repo projection is REFUTED by the audit. | none | [maxgap-law.md](maxgap-law.md) |\n| `Q-maximal-law-tpc` | ANSWERED | Is the SIEVE Gaussian maximal law TPC-implying? | YES in the only form that would deliver anything: stated for the operative remainder R_H, need/z^2 runs 0.49 to 0.61, flat and below the zone budget at every z from 13 to 43, which is the Gap Reformulation, hence the Zone Postulate, hence TPC, so no soft argument can prove it; the published above-2 theta column is not the maximal law's verdict but the law plus two lossy steps. | 0 | [phase1-T4-maximal-law.md](history/staging/phase1-T4-maximal-law.md) |\n| `Q-maxsum-growth-law` | CLOSED | What is the growth law of maxsum_m, and is there a constant C that closes the localized chain? | The chain has no chance, and for a reason cheaper than the sieve wall: the gate grows like ln x per fold while the bound it must stay under grows by at least 2.4 ln^2 x, a deficit that is unconditional in C and in the sifting parameter, and the Deficit Lemma kills the weaken-the-gate repair because the gate feeds back. | none | [localized-04-maxsum.md](localized-04-maxsum.md) |\n| `Q-maxvr-uniform` | CLOSED | Is there a level-uniform bound on max VR, and what would an all-x form need? | No level-uniform bound: all three candidate bounds are killed by the numbers, and the all-x form needs a second ingredient the Status sentence does not name, an analytic lower bound on Sbar, the count of n with both n and n+2 free of prime factors in (x, sqrt(N)]; a max VR bound is in any case not TPC-strength, and no REFUTED.md row covers the route. | none | [attack-maxvr-uniform.md](history/staging/attack-maxvr-uniform.md) |\n| `Q-measure-0904-argmax` | PARTIAL | Where inside the tile is G2(x#) attained, with what multiplicity, and how thick is the near-maximal tail? | The multiplicity half was NOT open, and is reproduced here rather than found: exact-g2-ladder.js, measure-g2z2-0829.js and ATTACKS3.md already carry it at every level to x = 43, so object-g2-read-0829.md:1000's \"unknown\" is wrong. What is new, MEASURED exhaustively over the whole period at x = 11..37 by a third independent engine: the argmax set is tau-invariant with no fixed-point hit at all eight levels and carries ZERO congruence pairs from x = 23 up (refuting this note's own F7 at four scored levels), while 58 to 82 percent of the near-maximal windows ARE exact congruence translates of another by a multiple of W/p, at 10^6 to 10^8 times the uniform null; at x = 37 the maximum is 2 positions of 7,420,738,134,810 and only 6 gaps of 217,929,355,875 sit within 10 percent of it; four of eight pre-registered forecasts failed (F3, F5, F6, F7) and two falsifiers fired (F5, F7), and the exponent is untouched. | none | [measure-0904-argmax.md](history/staging/measure-0904-argmax.md) |\n| `Q-measure-g2-generic` | ANSWERED | Where does G2(x#) sit in the distribution of the same maximum gap taken over other two-class configurations with the same census, and is the constant-shift family generic inside the free-pair family? | G2 is GENERIC in this coordinate on the registered rule, MEASURED: the midrank percentile of G2 in the census-matched two-class ensemble reads 20.00, 18.06, 26.35, 17.14 exhaustively at x = 11..19 and 13.70 sampled at x = 23, all inside the registered [5, 95] band so the falsifier does not fire, while sitting about one ensemble sd below the mean at all five levels; and the free-pair and constant-shift ensembles are PROVEN identical up to translation, so killer 1 gains no target in the class positions at these levels. | 0 | [measure-g2-generic-0829.md](history/staging/measure-g2-generic-0829.md) |\n| `Q-measure-g2z2` | ANSWERED | What is G2(x#)/Z2(x) at every level where both exist, where does the tile's least attaining position sit, and how often does a zone hold exactly two pairs? | MEASURED at 22 levels: G2/Z2 rises 1.00 to 8.1429 with its maximum 8.3214 at x = 71, is lossless only at x <= 7, and no gap of length G2 lies inside the zone for x = 11..79 (PROVEN there from Z2 < G2 plus the Zone Restriction Lemma); R0's tight case k = 2 occurs at 0 of 27,291 zones above the degenerate first one; the fourteen custody ratios reproduce digit for digit and Z2 = env at 27,292 of 27,292 zones; the pre-registration scored three of five with one pre-registered theorem refuted (the Poisson miss is 1.518e-1 over p >= 3, corrected from a mislabelled 4.177e-1); no upper bound on G2/Z2 is claimed or in evidence. | none | [measure-g2z2-0829.md](history/staging/measure-g2z2-0829.md) |\n| `Q-minus-half` | ANSWERED | Is the -1/2 anticorrelation constant of the abutting-window covariance exact, or only equidistribution-conditional? | For the flat spectral functional the -1/2 is EXACT, a two-line consequence of the mod-30 difference set, verified to relerr <= 1e-15 on the full spectrum at x = 7 and 11; the whole equidistribution question compresses exactly into two residue-class correlation sums, and the residual all-x statement reduces to G30_agg(x) < 1/2. | none | [natal-cap-26-minus-half.md](natal-cap-26-minus-half.md) |\n| `Q-mismatch-adjudication` | ANSWERED | What is the true value behind each of readings-traceability's 56 mismatches, and where does each repair land? | All 56 adjudicated with old and new recorded: 33 READING-FIXED, 5 OUTPUT-FIXED-REBOUND, 2 awaiting re-embed, 5 DOC-FIXED, 4 UNTRACED-MARKED and 7 HOLDS-AS-IS; nine were repaired by making a script produce a figure it had only described, and three of those reproductions turned up a second defect. | none | [mismatch-adjudication.md](history/staging/mismatch-adjudication.md) |\n| `Q-mobius-bv-derivation` | ANSWERED | Is there a published theorem statement of Bombieri-Vinogradov for the Mobius function at level T^(1/2)/log^B T with a maximum over coprime classes, and if not, does that estimate follow from published theorems by a derivation whose every hypothesis is checked? | No published theorem statement was located in five channels searched in the owning convention named in research/SEARCH-CONVENTIONS.md; the closest is Granville-Shao's published assertion that the result is known, which carries no proof locator for the Mobius case. The estimate is derived here from four numbered results of Koukoulopoulos, GSM 203 (Corollary 13.4, Theorem 26.2, Theorem 26.6, equation (26.3)) plus Vaughan's identity for mu, with B(A)=A+6 and an ineffective constant. This is a derivation from published theorems, not a refereed theorem, and it is read from the author's preliminary version of that book. | C | [mobius-bv-derivation.md](mobius-bv-derivation.md) |\n| `Q-mod30-tail` | ANSWERED | Are the BV survey's S1 (mod-30 tail) and S2 (fixed truncation and depth) provable as stated? | Both PROVEN at short-note grade (no second reader) once S2's freshness factor is corrected to one class per prime, which needs only P^-(m) >= q > q_i and so holds at every scour prime, not only in the prime regime; tail constant 2 ln 2 -> (ln 2)/2; nothing in the head. | 11 (retired) | [thm-mod30-tail.md](history/staging/thm-mod30-tail.md) |\n| `Q-monotonicity-sweep` | ANSWERED | Does anything in the corpus assume certificate validity is monotone in L? | One unsound artifact and one invalid inference: attack-beta2-04-loss-budget.js section 6 bisects on a predicate measured not upward-closed and is wrong at 3 of 5 levels, true first-crossings 30/72/132/174/210 against the reported 36/72/144/174/354, and redteam-DP1-certificate.js draws a global minimality conclusion from a two-point local check; corrected, worst-casing certifies within 1.00 to 2.00 of true G2 and the exponent penalty runs 0 to 0.27 and falls with x. | 3 (retired) | [monotonicity-sweep.md](history/staging/monotonicity-sweep.md) |\n| `Q-moving-cutoff-parity` | ANSWERED | Does a proof-level reading of the shifted-Mobius sieve give a valid, explicit error-tolerant arithmetic target, and does its divisor switching preserve the moving boundary? | CHECKED prior art: Murty--Vatwani already supplies the conditional squarefree/parity mechanism and fixed-residue reduction. Its printed p. 654 swap omits n+h>ey; an exact finite counterexample verifies failure of that equality. DERIVED dyadic repair keeps the boundary and gives S=C2*x-2*C2*M+D_y+O_A(x/log^A x). A one-sided D_y>=-4*x/25+o(x) on unbounded dyadic scales is sufficient, using the certified margin C2*(1-A2)>33/200. That discrepancy estimate remains OPEN. Vatwani's recovered author preprint clarifies the one-odd-exponent cases and unfilled near-level-one hypotheses. No new signed bound is obtained; literature novelty is unestablished. | C | [moving-cutoff-parity.md](moving-cutoff-parity.md) |\n| `Q-mp-derivation` | ANSWERED | Can the fold-factor field M_p be derived rather than merely measured? | Its comb structure derives: M_p is about k*W1(theta_p)*exp(-delta*theta_p/mbar_p) with W1 the exact endpoint comb at zero parameters, blind-validated at a second fresh anchor; (k, delta) remain fitted, so this is a zero-parameter derivation of the field's comb structure and not of M_p, and nothing here proves H-double-prime or touches the exponent. | none | [mp-derivation.md](history/staging/mp-derivation.md) |\n| `Q-mp-window-prereg` | OPEN | Do the M_p FORMULA bands hold at a second fresh anchor? | Pre-registration only, sealed and committed alone before the measuring producer exists: the window, the population, the zero-new-parameter formula, all 37 band pairs, three scoring criteria and the consequences are fixed here, pinned to the v1 producer's embed fingerprint; nothing is measured. | none | [mp-window-prereg.md](history/staging/mp-window-prereg.md) |\n| `Q-multiplicity3` | CLOSED | Does the m >= 3 multiplicity instrument buy anything for the anchored deficit? | Closed as an instrument: at @17 the m >= 3 statistic sees the anchored deficit worse than the pair statistic it was to replace (anchored z -0.23 against -2.71 for the overlap credit X), the Bonferroni depth ladder buys one place and dies, and multiplicity 3 in moment order buys a factor 1.3 that does not reach the anchor. | none | [attack-multiplicity3.md](history/staging/attack-multiplicity3.md) |\n| `Q-natal-cap-campaign` | ANSWERED | What is the mathematical upper limit on the number of Natal@5 slots a prime q can remove from an X-tile as the Scour drags across it? | Per-prime removal is boxed by four instruments (max overshoot +7.7% over the fair share 2N/q, worst conceivable dilation +0.7% at @17 over 92,160 tuples, certified Selberg caps 1.13-2.5x truth, and a proven prime-counting cap for q > W^(1/3)), but no per-prime cap, even an oracle, closes the first-order union bound for x >= 13 because Sum 2/q crosses 1 between @11 and @13. | none | [NATAL-CAP-CAMPAIGN.md](NATAL-CAP-CAMPAIGN.md) |\n| `Q-natal-cap-overlap-sign` | CLOSED | Is the sub-CRT overlap bound S2 <= sum of 4N/(q q') provable, per pair or in aggregate? | REFUTED as an inequality at every granularity, with 40 to 53 percent of pairs sitting above CRT at x = 11 through 23 and named exhibits at each level; what survives is proven structure, an exact per-pair-computable aggregate deficit, with the remainder a sign-alternating discrepancy at noise scale. | none | [natal-cap-12-overlap-sign.md](natal-cap-12-overlap-sign.md) |\n| `Q-natal-cap-sieve-cap` | ANSWERED | What is the rigorous sieve upper bound on how much a single scour prime can remove, and does it certify the zone? | Certified Selberg and Bonferroni caps run 1.60 to 2.36 times N over x = 7..17 with the fluctuation factor at 1.000, gross(q) hugging 2N/q to three decimals, but from x = 13 the harmonic factor alone exceeds 1, so Sum gross > N and even exact-truth caps fail. | none | [natal-cap-10-sieve-cap.md](natal-cap-10-sieve-cap.md) |\n| `Q-natal-discrepancy-lemma` | PARTIAL | Can the rotation ensemble's class discrepancy be controlled strongly enough to feed Freedman's inequality? | Lemmas 1 to 4 and Theorem 5 are stated and machine-checked (1,954 assertions, zero failures) and Proposition 6's stopped Markov-union-Freedman chain is rigorous, but none of it touches the anchored problem: the anchored path's per-step conditional variances are ensemble-typical (mean ratio 0.93 to 0.99), so the anchored side remains 100 per cent drift against the Mertens/2C2 wall. | none | [natal-cap-14-discrepancy-lemma.md](natal-cap-14-discrepancy-lemma.md) |\n| `Q-natal-onset` | ANSWERED | Does Chris's q^2 destruction-onset staircase hold, and does the onset lens open a mechanism at x = 37? | The staircase is VERIFIED over 1..10000 after aborting-on-mismatch against his three hand anchors, the three onsets collapse to two and the naive 50:50 A/B split is refuted by derivation; the x = 37 fourth lens is a clean negative and the fourth mechanism family closes. | none | [natal-onset-01.md](history/staging/natal-onset-01.md) |\n| `Q-next-correlation-source-map` | ANSWERED | Is there a primary-source statement with stronger scale or coefficient quantifiers than the ones the prime-band transfer actually consumed? | Scoped negative for the inspected statements: function-class and rate mismatches remain. The coordinating review withdraws the all-scales/dyadic-omitting requirement as a necessary gate, since bounded-interval stability yields a weak dyadic scale average from the existing input. No sufficient rate or full-coefficient theorem is matched. Proper-prime-power terms are already negligible on the fixed corner; other prime-r cofactor branches remain open. | C | [next-correlation-source-map.md](next-correlation-source-map.md) |\n| `Q-next-transfer-review` | ANSWERED | Does prime-band-transfer.md (1)-(6) follow at exactly its stated scope, and does any step fail against an adversarial reading of its hypotheses? | The derivation (1)-(6) survives at its stated continuous-average scope. The coordinating round review corrects check l: interval stability additionally gives a dyadic scale average, with a further exponent loss. There is no every-dyadic estimate, o(N), full-corner control or twin margin. Proper-prime-power branches have a separate negligible fixed-corner bound. | C | [next-transfer-review.md](next-transfer-review.md) |\n| `Q-novelty-postaudit` | ANSWERED | Of the ten results never covered by the 2026-08-13 prior-art audit, which are novel? | None survives as fully novel: two are outright prior art (one published thirteen years earlier with its headline claim contradicted by the source's own table), one is a two-line classical theorem mis-calibrated as a measurement, five are restatements or special cases of standard objects, and one names a lemma that does not exist. | none | [audit-novelty-postaudit.md](history/staging/audit-novelty-postaudit.md) |\n| `Q-np-licenses` | ANSWERED | What does NP-completeness license here, and is Brady's Problem 3 the G2 covering problem? | The corpus sentence \"his Problem 3 IS the G2 covering problem\" is false as written and must be withdrawn (one class per prime and an arbitrary finite set, against two classes at locked separation 2 on an interval); NP-completeness licenses four statements about computation and none of the five mathematical ones. | none | [attack-np-licenses.md](history/staging/attack-np-licenses.md) |\n| `Q-null-limsup` | MIXED (null-limsup-prereg.md: OPEN; null-limsup.md: ANSWERED) | Does the thinning null's maximal gap disagree with Neudecker's limsup law, which would retire the extinction half? | The trigger does not fire: the closed form is the exact law, the reduction returns Hawkins' own constant, and the real fold's record gap lands within 2 sigma of the null, so the extinction half survives and the null now supports the ceiling argument rather than threatening it; the limsup itself is not measurable and cannot be. | none | [null-limsup-prereg.md](history/staging/null-limsup-prereg.md), [null-limsup.md](history/staging/null-limsup.md) |\n| `Q-object-bridge-read` | PARTIAL | What sits between Z2(p) and G2(x#), which inequalities between them are proven and in which direction, and is the information the Gap Reformulation discards provably needed? | Map only, nothing computed: the reduction discards the origin to buy rotation invariance at a loss measured at G2/Z2 = 4.12 by p = 43, and the discarded information is PROVABLY not needed for the implication (a rotation-invariant statement, G2 < x'^2 - 2, does reach the zone statement), MEASURED to be worth about one unit of exponent against a deficit of 2.2665, and OPEN as to whether any proof needs it, with no separation theorem on record on either side. | none | [object-bridge-read-0829.md](history/staging/object-bridge-read-0829.md) |\n| `Q-object-g2-read` | PARTIAL | What is actually known about G2, at what calibration, where do the corpus's documents disagree about it, and which questions about it have not been asked? | The band is unchanged: exponent 4.26645 proven above, 2 needed, and the reading found no document that moves it; what this note adds is a variant table, a legal open set for section 4, seven drift items including one live target that is TPC-strength and unlabelled, and eight unasked understanding questions. | none | [object-g2-read-0829.md](history/staging/object-g2-read-0829.md) |\n| `Q-object-models-read` | PARTIAL | For G2(x#) and Z2(p), what does each null model in the corpus predict, which measured deviations from those models are real, and does any of them carry interval-level (wall-relevant) information? | Nothing found that carries interval-level information: seven of the nine real deviations are residue-level by the THE-LENS section 5 rule, Kourbatov's b is interval-level in form but blocked by a log-sparse quantifier and unexplained as a mechanism in both corpora, and the sub-Poisson rough-pair dispersion is classified interval-level in form and elementary in content (division with remainder, not placement) by the independent-thinning null that ran 2026-08-29 (measure-roughpair-null-0829.md), which also weakened the sub-Poisson reading since that null's own chi-squared per degree of freedom is 1 minus p rather than 1. | none | [object-models-read-0829.md](history/staging/object-models-read-0829.md) |\n| `Q-observations` | PARTIAL | Which sightings from the bench survive being stated precisely enough to check, and at what rung? | A running log, each entry carrying its own rung: the mirror-centre and divisor-landmark items stand as VERIFIED, the closed form for the half word is REFUTED as a route (max of the half equals max of the whole at every level), and the compounding window is PROVEN to close only to U = 30. | none | [OBSERVATIONS.md](OBSERVATIONS.md) |\n| `Q-obstruction-audit` | ANSWERED | Is the wall named parity the right name for what blocks G2(x#) < x'^2 - 2? | Read at source, the audit's central reason does not survive: Tao's forbidden is defined extensionally rather than intensionally, so the primality-detection step in the reduction does not put the method class outside the obstruction, and where the wall sits on our own reduction moves accordingly; the proposed text for paper/wall-note.md is left unapplied. | none | [attack-obstruction-audit.md](history/staging/attack-obstruction-audit.md), [lit-tao-parity.md](history/staging/lit-tao-parity.md) |\n| `Q-oeis-G2-submission` | CLOSED | Should the twin Jacobsthal function G2 be submitted to OEIS as a new sequence? | DO NOT SUBMIT: it duplicates A144311 (Carter, September 2008), which is G2 - 1 on the same fixed classes with no free translate and carries 22 terms to our fourteen, all fourteen agreeing; five waves missed it because every search was run on G2 and never on G2 - 1. | none | [oeis-G2-submission.md](oeis-G2-submission.md) |\n| `Q-oeis-g2-absence` | ANSWERED | Is the G2 ladder absent from OEIS at fourteen terms? | Falsified: the ladder has been in OEIS since 2008 in the run-length convention G2 - 1, as A144311 carrying 22 terms to our 14, with the search channel calibrated live in the same session on two known-positive probes. | none | [audit-oeis-14term.md](history/staging/audit-oeis-14term.md) |\n| `Q-oeis-seam-submission` | PARTIAL | Is the seam twin-pair count sequence absent from OEIS, and is the draft ready to submit? | Status DRAFT: the exact-terms absence survives on both range conventions and the 20 terms are cross-checked by two independent Miller-Rabin implementations, but qc-wave6-X section B3 found a concept neighbour, A367739, that the draft's crossrefs do not mention. | none | [oeis-seam-submission.md](oeis-seam-submission.md) |\n| `Q-ojaroudi-read` | ANSWERED | Does Ojaroudi's Zenodo preprint prove an unconditional twin-prime theorem, and is anything in it importable? | No: the fatal step imports a Selberg quadratic-form LOWER bound from Opera de Cribro Chapter 7 where none exists, and a second independent gap leaves the Kloosterman hypothesis never checked; the venue is an unrefereed self-deposit, and the only payoff is two elementary lemmas that are importable onto our tile. | none | [ojaroudi-read.md](history/staging/ojaroudi-read.md) |\n| `Q-omega-floor-blind-0830` | ANSWERED | Does the local log-slope of the pointwise floor Omega(z) of attack-0830-rec-cheapest.md (exact to z = 73, certified 2+4-chain family to z = 5e5), extended blind to new levels under a sealed forecast, behave as the DERIVED growth law Omega >> z^{16s/9}/ln^8 z predicts at s = 2.698721, and at what z would the test be decisive? | MEASURED, blind, on the derived law only: 18 of 22 sealed rows HIT; the certified 2+4-chain family at p* = (47, 43) runs 5e5 -> 1e9 (A1A2 = 8.863e17 -> 9.351e32, exact integers) with its step slope falling 4.85 -> 4.38 through every sealed band onto the law's log-corrected slope, the decisive decade 1e8 -> 1e9 reading 4.410 against 16s/9 - 8/ln z = 4.389 (HIT, kill rule not triggered at 4.452 +/- 0.011); the 4 MISSes are the ratio to z^{16s/9}/ln^8 z sitting at 2.00-2.14 above the sealed cap of 2 at z >= 1e8, the pre-declared non-adverse direction; the exact floor is extended 73 -> 113 (251 -> 1980, equal to the hill-climb floors at 89 and 101), where the law predicts nothing; no exponent moves and the law stays DERIVED and HELD. | 0 | [blind-0830-omega-floor.md](history/staging/blind-0830-omega-floor.md) |\n| `Q-operator-pair-count` | ANSWERED | Can the histogram transfer operator and the pair count be evaluated exactly at any level with no tile in memory? | Both instruments are exact and neither needs a tile resident: the operator is PROVEN and VERIFIED exactly at six folds and the pair count at every fold; the tail carries a FIT beside them that must not be used past the range it was fitted on. | none | [operator-and-pair-count.md](operator-and-pair-count.md) |\n| `Q-origin-excess` | ANSWERED | Can the Origin Excess Lemma be pushed to S = x'^2, and is the scale collision genuine or an artifact of that one argument? | It cannot, and the collision is genuine: the origin's excess is exactly a ratio of two values of the survival curve rho at the same window and two sieving levels, and the Zone Postulate asks for that ratio at u = 2, where rho takes its minimum, so no better lemma moves the boundary; three candidate mechanisms fail, each for a different reason, one of them working against us. | none | [origin-excess.md](origin-excess.md) |\n| `Q-packing-two-class-priorart` | PARTIAL | Does the two-class (twin) analogue of the Hensley-Richards packing function exist in print? | Not found, on an uncalibrated search in our own wording with no owning convention recorded, so this is the reach of the search rather than an absence; the classical one-class theory is well developed, and against A008407 the two-class ladder is bounded below by the minimal admissible 2k-tuple diameter. | none | [natal-cap-04-packing-notes.md](natal-cap-04-packing-notes.md) |\n| `Q-paired-factor-budget` | ANSWERED | Does averaging the proposed two-small-prime loss and three-small-prime gain using existing distribution theorems make their net contribution sufficient for the current twin consumer? | DERIVED from the matched classical inputs, with finite rational enclosures VERIFIED: the explicit positive family has liminf P3/(C2*x)>1, but liminf (N2-P3)/(C2*x)>11/10. Keeping only these families and the main term therefore fails at the current cutoffs. Pan--Ding supplies aggregate level x^(9/20), stronger than the earlier per-cofactor application; the complete signed margin remains OPEN. The aggregation machinery is prior art and literature novelty of this application is unestablished. | C | [paired-factor-budget.md](paired-factor-budget.md) |\n| `Q-paper-iii-restructure` | ANSWERED | Is Paper III restructured around the proven model theorem? | Yes. paper/variance-note.md now carries four new sections in place of its fit narrative: sec.8 the proven spectral form (Var/E = delta X, delta ln^2 W -> 16 C_2 e^{-2gamma}/3 = 1.109905, the Montgomery-Soundararajan main term exactly L prime by prime, so all of Var/E is discrepancy), sec.9 the decoupled model with Theorems 7 and 8 and Lemmas 9-11 written out with proofs and the closed form lambda_2(2) = 1 - e^{-2gamma}(9/2 - 4 ln 2) = 0.45546 plus the theta = 1 identity with Gorodetsky, sec.10 Conjecture 1 (delta(X - X_dec) -> 0) with eight exact ratios, two of three lag groups closed and the third named as a divisor-distribution statement not yet searched, and sec.11 the 0.611 refutation as a short calibration paragraph; sections 1-5 are untouched, the abstract and section 6's open question are rewritten to match, and the \"u = 2 exactly\" defect in section 7 is corrected. | 9 | [paper-iii-restructure.md](history/staging/paper-iii-restructure.md) |\n| `Q-paper-kk-draft` | ANSWERED | Is the K–K two-class lower bound written up as a refereeable draft? | Yes. paper/kk-lower-bound.md, self-contained at INFERRED grade, carrying Theorem B (the substitution), Theorem A (the published-ingredient x ln x chain) and Proposition 2 (the free G2 >= g transfer), with every constant, every consumed hypothesis, the y0 = 10^134.1 floor, the three residuals, and a per-number provenance appendix; nothing in it is new mathematics and nothing in it is refereed. | none | [paper-kk-draft.md](history/staging/paper-kk-draft.md) |\n| `Q-parity-adversary` | ANSWERED | Does a parity adversary built on our own stretches survive at every level, and what precision class does a certificate need? | It does NOT survive: the exact-data adversary dies at D* about Q^1.18 (MEASURED, exact arithmetic, 43 anchors Q <= 200), and the certificate that kills it is itself destroyed by 0.30 of one count per modulus, so Z2's precision class is now measured rather than described; it is PROVEN not to be a proof ingredient. | Z2 | [attack-parity-adversary.md](history/staging/attack-parity-adversary.md) |\n| `Q-parity-dstar-law-0829n` | CLOSED | Does the exact-data parity adversary's killing level D*(Q) follow D* ~ Q^c with a stable c on the decades past Q = 200, and is c distinguishable from the independent-thinning null's? | No law in Q and CLOSED by the sealed kill rule (score 5 HIT, 1 MISS): MEASURED on 271 anchors 211 <= Q <= 13679, the Q-slope moves 1.446 to 1.204 to 0.575 across ranges while ln D* on ln width holds slope 1.097 +- 0.012 with a quarter of the scatter, D*/width about 0.5, and the independent-thinning null reproduces the width law within 0.025 in slope; the mechanism is that D* is where a modulus first reads a single position of the window, which is the wall survey's remainder statement in the stretch coordinate and names nothing new. | Z2 | [attack-0829n-parity-dstar.md](history/staging/attack-0829n-parity-dstar.md) |\n| `Q-perfold-error-model` | ANSWERED | What is the per-fold error model behind the extinction law's +/-3 sqrt(lambda) failure at 43.2 per cent? | X_p ~ Poisson(lambda_model(p, Y) M_p) with M_p a deterministic factor of the fold, the same number at every window length from 2e7 to 2e11 and at every anchor, MEASURED and not derived, and it holds on a sealed blind test at a fresh anchor. | none | [perfold-error-model.md](history/staging/perfold-error-model.md) |\n| `Q-perfold-window` | OPEN | Does the negative-binomial per-fold error model's band hold at a fresh blind anchor, after the extinction law's +/-3 sqrt(lambda) clause failed at the fifth window? | Pre-registration only, committed alone before the producer existed: the population is fixed in advance at the 37 folds p >= 100 with predictive mean or lambda_model at least 1, and the bands are printed verbatim from the model's embedded Stage 4; nothing here is measured. | none | [perfold-window-prereg.md](history/staging/perfold-window-prereg.md) |\n| `Q-phase1-W1b` | ANSWERED | Which unapplied wave-6 findings on the research notes were still live, and what did applying them change? | Most of wave 6 was already applied, so the yield is the residue plus one code defect nobody had fixed: a3-05-bound-L.js's truncated cyclic replay was printing a FALSE PASS on kappa(m) <= L+2 at fold 7, and fixing it kills a claim; the gate reads 0 findings across all seven checks at hand-back. | none | [phase1-W1b-applied.md](history/staging/phase1-W1b-applied.md) |\n| `Q-polylog-fold-transfer` | ANSWERED | Can the fixed-fold calculation be made uniform at growing depth, and transferred to the actual bilinear remainder with an adequate error? | The local bilinear model is O_H(x/log^H x) uniformly for polylogarithmic squarefree moduli. Its absolute reconstruction error is at least (1/2+o(1))*x. The needed signed reconstruction bound remains OPEN, so this model does not establish the twin-prime lower bound. | C | [polylog-fold-transfer.md](polylog-fold-transfer.md) |\n| `Q-prime-band-completion` | ANSWERED | Can the remaining squarefree correlation be localized to simpler prime and harmonic bands, and does completion then permit a direct general bilinear Kloosterman estimate? | The pilot reduces, with power-saving endpoint errors, to p in (x^0.235,x^0.24], q in (x^0.045,x^0.05] and positive h in (x^0.029,T], with T=O(x^0.0342). Both primes exceed 2h eventually. Completion is exact but produces coupled h-dependent coefficients on every dual residue modulo q. Its full-frequency Kloosterman matrix has norm exactly q, so a uniform power saving by an arbitrary-coefficient operator bound is impossible at that scope. This completion alone leaves the 61/20000 budget deficit; prime-dispersion.md controls the pilot using a different second moment. No twin lower bound is established. | C | [prime-band-completion.md](prime-band-completion.md) |\n| `Q-prime-band-transfer` | PARTIAL | Can the nonmultiplicative prime-band weight be represented by bounded multiplicative functions, and what does an available quantitative correlation theorem actually give after all scale and normalization costs? | A smooth Fourier superposition represents mu(n)L_B(n)/log X exactly on n<=2X+2. The quantitative non-pretentious input and a mesh give a continuous scale-average saving for the exact s=s'=1 windows. The round review additionally derives a dyadic scale-average saving by bounded-interval stability, with exponent c_*/2. Neither estimate gives an every-dyadic bound, o(N), full-corner control or a twin margin. The extension in cofactor-progression-transfer.md now estimates a growing polylogarithmic cofactor family; the full remainder remains open. Proper-prime-power branches have the separate bound in corner-correlation §1.1. | C | [prime-band-transfer.md](prime-band-transfer.md) |\n| `Q-prime-detection-inputs` | ANSWERED | Can we give a locally correct positive comparison sequence, justify the ordinary arithmetic inputs, and prove exactly what one additional estimate would buy? | Yes: the comparison sequence and Type I estimates are derived, and a classical Vaughan decomposition gives S(x)=C2*x+B(x)+O_H(x/log^H x) for every fixed H. A fixed one-sided saving over B>=-C2*x+o(x) implies twins on unbounded scales. This is a fully specified classical reduction, not a new cancellation estimate or proof mechanism. | C | [prime-detection-spec.md](prime-detection-spec.md) |\n| `Q-prime-dispersion` | ANSWERED | Can a second moment retaining two right primes control the remaining localized squarefree pilot after its actual conductor and coefficient costs are included? | Yes at this rectangle: aggregation in m=dp before Cauchy gives a logarithmically bounded coefficient on a long interval. The only zero-phase pairs are identical (q,h). Distinct primes give modulus e*q1*q2 with gcd loss at most O(A*Z); completion and Weil bound all off-diagonal terms. The diagonal has exponent 1989/2000 and the largest off-diagonal exponent is 1173/1250 before an arbitrarily small loss. The remaining high-frequency polynomial is O(x^(199/200)); consuming the previous reductions gives R_IJ=O_H(x/log^H x) at D=x^0.277, E=x^0.467. This is a full rectangle at product scale x^0.744, not a uniform product cutoff or a twin lower bound. | C | [prime-dispersion.md](prime-dispersion.md) |\n| `Q-prime-power-dispersion` | ANSWERED | Does the repeated-prime structure control the BB contribution beyond its two-norm bound, and can a prime-power moment include every harmonic uniformly? | The decomposition leaves an exceptional coefficient with norm at most sqrt(D) times logarithms; its odd prime-power core has a derived all-harmonic dispersion bound. The same bound for first powers removes the former left-prime cutoff and unpaired diagonal constraint. Two strict inequalities control a region including d~x^0.247, e~x^0.612 at product scale x^0.859. Their product-exponent supremum is 5757/6700, not a uniform product cutoff or a twin lower bound. Exact finite checks run separately from this classical-input derivation. | C | [prime-power-dispersion.md](prime-power-dispersion.md) |\n| `Q-prior-art-audit` | PARTIAL | What in this programme is rediscovery, and what might be new? | Rediscovery, prolific and correct, spanning 170 years (Smith 1857 to Tafula 2015); three candidate novelties survived the 2026-08-13 audit and item 1 was cut back on 2026-08-18 by OEIS A144311 (Carter 2008), and several negatives were run in our own wording rather than the owning convention, which is worth little. The 2026-09-06 global smooth majorant uses an explicitly matched prior-art finite-difference mechanism; literature novelty of the complete application is unestablished. | none | [PRIOR-ART.md](PRIOR-ART.md) |\n| `Q-prior-art-last-ground` | ANSWERED | Is MathSciNet reachable, and does the two-class upper-bound absence survive a complete title enumeration? | MR Lookup answers author/title/journal/year queries free (bibliographic fields only, no review text or MSC), so SEARCH-CONVENTIONS section 5 needs correcting; the standing sentence survives, since the complete MathSciNet title record for Jacobsthal contains every published upper bound and every one is one class per prime, and Holt's 2022 book is in no bibliographic database. | none | [attack-prior-art-last-ground.md](history/staging/attack-prior-art-last-ground.md) |\n| `Q-proposals-prior-art` | ANSWERED | What is the prior-art risk for the six candidates in the paper-proposals registry? | None is clean: three are ADJACENT with the boundary drawn Holt-style (one of them breaking a sentence the corpus rests on), the thinning null is PRIOR ART FOUND under Hawkins' random sieve with the Fold Moment Identity NOT FOUND, and only the reverse-direction Suen import reads NOVEL-SO-FAR; every verdict names what was not searched. | none | [proposals-prior-art.md](history/staging/proposals-prior-art.md) |\n| `Q-qc-arch` | ANSWERED | Is the corpus's architecture, ordering and file layout readable by a fresh agent at minimum context cost? | Diagnosis only, no existing file edited: the restructure plan of section 2 was accepted over the shepherd's, and the leading verdict is to leave OBSERVATIONS.md exactly as it is, in place, unsplit and off the reading path, taking out only four inbound citations that name it as the home of something it does not own. | none | [qc-arch.md](history/staging/qc-arch.md) |\n| `Q-qc-campaign` | ANSWERED | What did the 2026-08-17 consistency campaign run, find, decide and apply? | The shepherd's ledger: Chris's duplication rule and splitting ruling recorded in his words, the qc framework and the provenance index delivered in wave 1, and every partition's results with the decisions that settled them, including two novelty verdicts flagged as publication-track changes. | none | [qc-CAMPAIGN.md](history/staging/qc-CAMPAIGN.md) |\n| `Q-qc-checks-9-10` | ANSWERED | What do the provenance and search-convention gate checks find, and why are they held out of the default gate? | Both are built, calibrated and in the selftest but deliberately behind --pending because they fire on live claims an attack agent is mid-flight against: provenance reads 8 findings and search-convention 75 on the 2026-08-18 corpus, the flip to default is one line, and what could not be mechanised is stated. | none | [qc-checks-9-10.md](history/staging/qc-checks-9-10.md) |\n| `Q-qc-compound` | ANSWERED | Which named claims are compounds whose parts carry different calibrations, and how should each be decomposed? | Twelve compounds decomposed and eight examined and left whole, with about 90 corpus sites needing rewording; the test each decomposition must pass is that Proven: becomes unwritable rather than merely wrong, and the sharpest finding is that natal-cap-28's fully analytic certificate law has no markdown home at all, which is why its status is unfindable. | none | [qc-compound.md](history/staging/qc-compound.md) |\n| `Q-qc-history-migration` | ANSWERED | Which body text is history and should move to CHANGELOG.md or history/? | The headline is a negative and it is the most useful thing in the report: no body file is a session record that should move wholesale, so what moves is per-section, and the pass leaves an ORPHANED-CLAIM list of live errors plus per-file findings over all 63 body files. | none | [qc-history.md](history/staging/qc-history.md) |\n| `Q-qc-numbers` | ANSWERED | Do two places in the corpus ever give the same number for the same object, and does each remote quotation carry the caveat its home document requires? | 16 findings organised by quantity, led by the rho disagreement between U-FRAME:646 and gate-multiplies:360; diagnosis only, no existing file touched, and arithmetic-per-number is audit-numbers.md's axis rather than this pass's. | none | [qc-numbers.md](history/staging/qc-numbers.md) |\n| `Q-qc-papers2` | ANSWERED | Do the paper suite's five doors and four faces restate their research homes correctly? | Fifteen findings in sections 7 and 7A plus a suite sweep: a door resting entirely on a column its own cited file disclaims as invalid, whose corrected computation reverses the reading; a degree-4 certificate credited with a result needing moments through order 6; and the same paper contradicting itself forty lines apart on the anchored calm. | none | [qc-papers2.md](history/staging/qc-papers2.md) |\n| `Q-qc-queue-9-10` | ANSWERED | Can the provenance and search-convention queues be cleared without suppressing anything? | TOTAL 84 to 0 with nothing suppressed and every finding resolved in the corpus: all six not in OEIS claims were FALSE, since the G2 ladder is A144311 + 1 and PRIOR-ART.md already recorded the find 270 lines further down; the honest remainder is that five of 75 were resolved by rewording a sentence that was never a literature claim. | none | [qc-queue-9-10-worked.md](history/staging/qc-queue-9-10-worked.md) |\n| `Q-qc-refs` | ANSWERED | Are the corpus's external citations and its quotations sound? | Nine dead quotations and eight wrong external citations, the worst being that the corpus sits on the retracted FKMPT constant, an OEIS crossref off by one twice in a document addressed to an OEIS editor, and a false premise (\"Kanold, Stevens and Paseman all give exponent 2 + eps\") under an open TODO item. | none | [qc-refs.md](history/staging/qc-refs.md) |\n| `Q-qc-scope-R` | ANSWERED | Where does the summary layer restate a result with the scope its home document attaches dropped? | Fifteen findings, 2 HIGH, 3 MEDIUM-HIGH, 4 MEDIUM, 6 LOW: one HIGH is a false statement in README.md Status introduced by this campaign's own wave 2, the other is both papers giving a false reason for the X-limitation theorem's scope, and the coverage statement at the end is short and matters. | none | [qc-scope-R.md](history/staging/qc-scope-R.md) |\n| `Q-qc-scope-T` | ANSWERED | Does the surface partition R left unswept carry the campaign's dominant defect, a scope dropped in restatement, or a false absence claim? | Fifteen ranked findings plus the TA, TB and TC blocks: a scoreboard still logging the rich vein as the campaign's one unexplained novelty after a same-day follow-up solved it, a fourth copy of the retired sifting limit, and a largest-single-event superlative refuted by the event log it is read from; adjudication only, nothing edited. | none | [qc-scope-T.md](history/staging/qc-scope-T.md) |\n| `Q-qc-scripts-S1` | ANSWERED | Do the 37 natal-cap scripts' READINGS blocks hold against their own output and against the documents citing them? | Twenty-six of 37 are consistent and eleven carry a defect, none of them contradicting the summary layer on a status claim: a false promise of a record with no READINGS block, a reading refuted by its own printed output, an absence claim at the site a whole wave read as absolute, and a Miller-Rabin bound wrong by nine orders (3.186e14 for 3.186e23) at three sites. | none | [qc-scripts-S1.md](history/staging/qc-scripts-S1.md) |\n| `Q-qc-scripts-S2` | ANSWERED | Do the fold-profile and attack script families say what their own outputs say? | Thirteen defects over 34 scripts and 5,673 lines, five of them HIGH: an untested identification two later scripts had already killed, a printed reading refuted by its own table, a section that has printed only its header since it was written, and a reproduction pointer to a scratchpad that is gone; script-side ones fixed, document-side ones left as evidence, gate 0 on all seven checks. | none | [qc-scripts-S2.md](history/staging/qc-scripts-S2.md) |\n| `Q-qc-scripts-S3` | ANSWERED | Do the 57 non-campaign research scripts' banners, embedded OUTPUT blocks and READINGS agree with each other and with the documents citing them? | Defects ranked by consequence, the leading one a3-02-diagonal-f.js carrying the refuted L scanner while naming as its source the sibling that had been repaired underneath it, the two state machines disagreeing at exactly one cell, L(T23, 29) = 3 against the true 2, now fixed; the partition ends with qc.js at 0 on all seven checks and audit-numbers.js 78/78. | none | [qc-scripts-S3.md](history/staging/qc-scripts-S3.md) |\n| `Q-qc-shepherd` | ANSWERED | What does the wave-1 shepherd find on the corpus's own structure? | The largest structural defect is that U-FRAME is organised by campaign rather than by claim, with five whole sections of process record sitting in working documents, a PROVEN lemma quoted without the hypothesis that makes it non-vacuous, and three documents disagreeing about how many hypotheses that lemma has; the claim-status register across named objects was not covered and is named as the highest-value remaining axis. | none | [qc-shepherd.md](history/staging/qc-shepherd.md) |\n| `Q-qc-slopes-K` | ANSWERED | Are the three fits at slopes 1.062, 1.451 and 1.2992 numerically inconsistent? | There is no numerical inconsistency: the seven tile points are seven of the 42 census points with f agreeing to printed precision, restricting the 42 to the seven's x reproduces 1.062 exactly, and the two estimates are statistically indistinguishable (seven-point 95 per cent interval [0.32, 1.81]); but the corpus's stated reason is wrong, the comb does not explain it, and the coefficient is range-dependent under every specification, so no single slope exists. | none | [qc-slopes-K.md](history/staging/qc-slopes-K.md) |\n| `Q-qc-status` | ANSWERED | Is any named object's claimed status inflated, contradicted or misattributed? | Thirteen findings, 12 with replacement text: the worst scope inflation is the Fused-Window Calm Lemma listed as proven when two of its four legs are not, and \"the matching lower bound\" claims a sharpness the corpus explicitly disclaims; on attribution the corpus is in better shape than the brief assumed. | none | [qc-status.md](history/staging/qc-status.md) |\n| `Q-qc-transfers-J` | ANSWERED | Are the thirteen transfers findings real defects or benign restatements? | Adjudicated by word diff with the diff pasted under each finding: 11 KEEP, 2 DEFECT, both in gate-multiplies.md and both one root cause, 0 false positives, and nothing needing compute or a human decision. | none | [qc-transfers-J.md](history/staging/qc-transfers-J.md) |\n| `Q-qc-wave6-U` | ANSWERED | Do ATTACKS.md and ATTACKS2.md say what the corpus above them says? | Fifteen findings on 57 lines, led by a campaign verdict awarding theorem-grade to a curve its own home marks MEASURED and Hardy-Littlewood-conditional, the rich vein still logged as an unexplained novelty, and the anchored-window programme called unexplored where a research note and a 34 kB paper exist; six of the fifteen turn on a frozen-record-versus-current-state decision nobody has made. | none | [qc-wave6-U.md](history/staging/qc-wave6-U.md) |\n| `Q-qc-wave6-V` | ANSWERED | Do the sixteen fold-profile scripts carry auditable evidence, and what do they show when run? | All sixteen were run and now carry a real OUTPUT block and a numbered READINGS block, each verified by re-running and diffing; eighteen findings follow, the highest being a lemma quoted as proven that is false as written in two documents while three scripts in this partition print the counter-example. | none | [qc-wave6-V.md](history/staging/qc-wave6-V.md) |\n| `Q-qc-wave6-W` | ANSWERED | Do the three expensive natal-cap artifacts reproduce? | All three re-run, after the complete @41 march log turned up on disk outside the repo in a place no wave had cited; a three-way tally per file, with the .md disagreements logged rather than edited and the coverage limits stated. | none | [qc-wave6-W.md](history/staging/qc-wave6-W.md) |\n| `Q-qc-wave6-Y` | ANSWERED | Do the eight run logs, the three committed .txt outputs and the leaf-note residue say what the corpus above them says? | Findings on all three surfaces, including a stated bound on sup\\|R\\|/(H*M) that the log's own full-period block violates by 5.6x, an S6 block produced by a function the repo has since flagged wrong while the log stays unflagged, and a false absence claim in G2-STATE that the same document refutes 162 lines later; line numbers were taken against a tree being modified concurrently. | none | [qc-wave6-Y.md](history/staging/qc-wave6-Y.md) |\n| `Q-qc-wave6-absence` | ANSWERED | Do the corpus's [ABSENT], [INFERRED] and [UNVERIFIED] markers survive a check against fetched primary sources? | First web-enabled pass, no repository file edited: the exact-terms OEIS absence for the seam count sequence survives on both range conventions but a concept neighbour, A367739, goes unmentioned in the draft's crossrefs, and the note closes with an explicit list of what it could not reach, led by the MathSciNet citation graph of Iwaniec 1978. | none | [qc-wave6-X.md](history/staging/qc-wave6-X.md) |\n| `Q-quadpoint-blind-31607-0830` | ANSWERED | Does the zero-parameter depth law y*(Q), stated on the set where T(Q) >= 1 and on both the 1/(2e^gamma) = 0.280730 and the u*omega(u) = 2 (1/u* = 0.280438) forms with and without the finite-size Mertens factor, survive a sealed blind test on the bands past Q = 10007, and can that decade tell the two asymptotes apart? | MEASURED, blind, and it decides nothing asymptotic: on 8363 new anchors 10009 <= Q <= 100003 the 76 sealed rows score 64 HIT and 12 MISS with every MISS on the two naked asymptotes as pre-declared, the two finite-size zero-parameter comparators HIT every row at every band (residual on ln y*/ln h -0.00022 to -0.00002 against C2), K*/pool falls 0.032 -> 0.026 -> 0.020 -> 0.016 -> 0.013 with no bend and a quiet sup, and the band means sit 0.0017 to 0.0037 below both asymptotes, six to twelve times their 0.000291 difference, so which limit the deficit tends to is set by the comparator and not by the data. | Z2 | [blind-0830-quadpoint-31607.md](history/staging/blind-0830-quadpoint-31607.md) |\n| `Q-quadpoint-c1c2-0830` | ANSWERED | Does the fourth decade's post-hoc, unsealed preference for the omega-form comparator C2 over the exact-Mertens comparator C1 (C1 above the data by 3.46 and 8.83 se at B11 and B12, C2 within 2.82 se) survive a sealed blind test at the fifth decade, Q in (100003, 316243], and what does that say about which mechanism explains the depth law's drift? | MEASURED, sealed before the run and scored blind on 17702 new anchors, and it decides nothing asymptotic: the sealed D1 row HITs at both B13 and B14 (C1 above the data by 13.33 and 23.82 se, C2 within 1.26 and 1.82 se), so by the sealed rule the omega-at-finite-u crossing of import-rough-anatomy.md §2.1 is CONFIRMED over the exact-Mertens product of u2-engine-depth.md §5 as the depth law's main term at the sampling scale, while both still HIT every ±0.0015 row, the kill rows on C2 all HIT, K*/pool falls 0.013 -> 0.010 -> 0.008 with no bend, and the band means sit 0.0012 to 0.0018 below both asymptotes, so the two constants 0.280438 and 0.280730 stay unseparated by the pre-declared rule and the run separates two mechanisms, not two limits. | Z2 | [blind-0830-quadpoint-c1c2.md](history/staging/blind-0830-quadpoint-c1c2.md) |\n| `Q-quadpoint-prior-art` | ANSWERED | In the owning conventions, who holds the rough-pair census, the capture identity and the crossing law, and is any of it ours? | The object and the identity are standard (the dimension-2 twin sifting function plus Buchstab's identity; ADJACENT-STANDARD, no verbatim carrier, no novelty language); only the comparison X(y) < T and the crossing law stay ABSENT-PER-CONVENTION, and the Buchstab correction is 6.6% of the top-band residual, not 2%. | Z2 | [quadpoint-prior-art.md](history/staging/quadpoint-prior-art.md) |\n| `Q-quadpoint-transplant` | ANSWERED | Does the anchored-cap family transplanted onto the stretch window S_Q certify every anchor, and what does the depth cost do with height? | Pre-registration for the decade extension, committed ALONE before the segmented producer existed: it fixes the FALL trend line, the registered band readings, the kill rule and a digit-exact calibration gate that must abort before any new figure prints, and states that a falling K*/pool over two decades would still be only a measurement. | Z2, Z1 (retired) | [attack-quadpoint-01.md](history/staging/attack-quadpoint-01.md), [attack-quadpoint-02.md](history/staging/attack-quadpoint-02.md), [quadpoint-decade-prereg.md](history/staging/quadpoint-decade-prereg.md) |\n| `Q-quartic-fourth-moment` | PARTIAL | Can the fourth moment of rho-tilde, attack-rhoms-01's named stopping object, be bounded? | Not yet: the genuine quartic correlation is small (OFF/m2^2 measured -0.0454, -0.2767, +0.0101, -0.0446 at z = 13..23), the resonance condition is proved to be a local covering condition, and the whole question transfers by two exact identities onto a Lambda(4) inequality for one explicit cosecant sum whose constant is flat; the new increment rung clears z = 47. | none | [attack-quartic-01.md](history/staging/attack-quartic-01.md) |\n| `Q-reachability-coverage` | ANSWERED | Which part of the original divisor domain can no improvement to the grouped moment's nonzero kernel reach, in either orientation, and what would a uniform saving on that kernel actually buy? | For a prescribed fixed exponent margin eta0, the zero budgets leave S_0={p<2eta0,q<2eta0} below that required margin regardless of the nonzero-kernel saving. Uniform gamma=2 covers its complement at that prescribed margin. Smaller positive margins can reach fixed interior points of S_0; only p=q=0 remains at zero budget one for every margin. The proposed eta0~loglog x/log x substitution is not justified by estimates carrying x^epsilon losses; its box count is geometry only. The sufficient twin margin remains OPEN. Independent review F3 and F5 correct the quantifiers and the prime-power tail proof; round-review-0906 corrects the diagonal scope, fixed-margin bulk claim and six-cut separation. | C | [reachability-coverage.md](reachability-coverage.md) |\n| `Q-readings-traceability` | ANSWERED | What is the source of each of the 1,592 figures that appear in a script's READINGS and not in its own OUTPUT block? | The running record: nothing in the pass changes a number, 56 mismatches are classified in eight kinds for adjudication elsewhere, and the advisory count rises when the work is done right because a provenance declaration is prose that carries printed values. | none | [readings-traceability.md](history/staging/readings-traceability.md) |\n| `Q-rec-cheapest-0830` | PARTIAL | At the cheapest legal point of REC(s, u0), s -> 1+sqrt(e) and u0 -> beta_2, where the saving asked over the triangle inequality is only z^{1.031}, does any of the five dead routes close, does a mixed mean-square-plus-cap strategy close, and is the failure a proof gap or a truth gap? | No route closes and the failing step is yesterday's statement, so PARTIAL on the routes; but the step is located as a TRUTH gap, not a proof gap, asymptotically: the certificate's own value at a CRT-planted doubly-smooth window is cc(r) = -(A1A2 + A1B2 + B1A2) with every term a count of exit chains of the Rosser supports (PROVEN, checked exhaustively at z <= 73), the window bound T <= cc(r) + H - 1 is exact, and the exit-chain count is DERIVED (held) to grow like z^{16s/9 - o(1)} for s <= 3, the hypothesis the 2026-08-30 red team added to the threshold, and z^{2s - o(1)} in the limit, so REC(s, u0) is false for every u0 < 16s/9 = 4.7088 at s -> 1+sqrt(e), the whole legal band (2, beta_2] included, and RML(alpha) with it, its falsity beyond z ~ 5.6e31; at every computable z it is a proof gap (Omega(z) exact to z = 73 reads 251, and the ledger's former comparator 1.05e6 is HM(u0), not z^{u0} = 7.188e7 at z = 73, so the gap is 68.3x larger than the ledger stated, while a certified lower bound 8.86e17 at z = 5e5 sits a factor 1.2e6 under z^{u0}); no exponent moved. | 0 | [attack-0830-rec-cheapest.md](history/staging/attack-0830-rec-cheapest.md) |\n| `Q-recon-0828-covering` | ANSWERED | Does the interval-covering, automated-proof or extremal-combinatorics literature hold a method that bounds the length of an interval covered by two residue classes per prime, or that turns the exact G₂ ladder into a statement at all x? | No route. Fifteen angles examined, fourteen dead on arrival or already closed here; the best interval-covering bound in print is Crittenden-Vanden Eynden's own Lemma 2, which reads 1.02e12 against G₂(79#) = 1710 and is 8181x weaker than the corpus's own 4.2665 sieve bound at the same level; the one survivor, transplanting Costello-Watts's recurrent certificate to two classes, is a computation that certifies finitely many levels and whose analytic half is Erdős #970 at one class. | 0 | [recon-0828-covering.md](history/staging/recon-0828-covering.md) |\n| `Q-recon-0828-farfields` | ANSWERED | Do any of the fields still absent from IMPORT-MAP.md hold a theorem whose hypothesis the tile satisfies exactly and whose conclusion is a lower bound on a periodic set's count in every window, a maximal-gap bound for a product-structured periodic set, or a positivity statement at a distinguished point? | No route from any of the sixteen candidates priced; two reach STRONG-ANALOGY or better and CLEAN and both bank only a closure or a wall address, the Fourier one by a one-line non-vanishing proof that empties the Donoho-Stark/Tao/Meshulam family on this object, and Janson's lower tail by saturating at exp(-1/c) >= 0.0805 against the 1/W a position union bound needs, short by 3.6e2 at x = 11 and 2.8e13 at x = 41. | 0 | [recon-0828-farfields.md](history/staging/recon-0828-farfields.md) |\n| `Q-recon-0828-jacobsthal` | ANSWERED | Does the 1978-2026 literature on upper bounds for Jacobsthal-type functions and maximal gaps in sieved sets hold anything that lowers G2 or Z2? | No route. Thirteen angles graded, ten dead on arrival (eight already closed in this corpus, two on their own structure), three surviving and none of them lowers an exponent: a kappa=2 extremal candidate built from an exceptional character rather than from Liouville, which can only close the band faster; an unread 1983 Siebert chapter that could already contain it; and one published anchor. Z2 has no non-TPC band at all. | none | [recon-0828-jacobsthal.md](history/staging/recon-0828-jacobsthal.md) |\n| `Q-recon-0828-rough` | ANSWERED | Does the 2000-2026 literature on y-rough numbers, sifted sets and short intervals hold any theorem that lowers Z2 or G2, meaning any control on the largest gap or the minimum count of a sifted set in EVERY short interval, or any structure of rough numbers a gap bound could use? | No. Ten angles examined, ten killed for this purpose; the strongest every-interval statement in print for one-class rough numbers is Iwaniec's S(x,y,z) >= (4y/log^2 y)(log(y/z^2) - O(1)) for y >> z^2, uniform in x, giving J(P(z)) << z^2, and it is the linear sieve run exactly at its sifting limit beta_1 = 2 with no slack outside it; no two-class analogue exists in print at any exponent, the corpus's own x^4.2665 is beta_2 by the same mechanism, and every remaining branch of the field (Matomaki-Radziwill, Gorodetsky, Friedlander-Iwaniec Cor 6.28, Gafni-Tao) carries the almost-all quantifier, which a second-moment statement can never convert into a bound on one empty window. | Z2, 0 | [recon-0828-rough.md](history/staging/recon-0828-rough.md) |\n| `Q-recon-0828-sieve` | ANSWERED | Is there anything in the 2008-2026 sieve literature, or in the parity-breaking literature, that lowers beta_2 = 4.26645 or supplies a consumer for a level theta > 1 on the two-class comb? | No route. Twenty-one angles examined, nineteen dead on arrival, nothing in print beats 4.26645 at kappa = 2 at any level on four calibrated channels; the two survivors are an owning-convention row and one unresolved hypothesis in a five-page 1990 announcement whose stated requirement beta < 3 excludes beta_2 = 4.26645. | 0, Z2 | [recon-0828-sieve.md](history/staging/recon-0828-sieve.md) |\n| `Q-recon-0829-escapes` | ANSWERED | Each of the three killers is an obstruction shape that occurs elsewhere in mathematics; where has each shape been beaten in print, by what mechanism, under what hypothesis, and does that hypothesis hold for the tile and zone objects? | Of nineteen escapes read, fourteen fail a hypothesis check (twelve on a fact already in this corpus), four fail some other way and one (the decoupled vector sieve) stays OPEN as the programme's existing road, so no route opens; Granville-Soundararajan's uncertainty principle is NOT killer 3 in print, is adverse-direction, and is already cited here twice. | 0 | [recon-0829-escapes.md](history/staging/recon-0829-escapes.md) |\n| `Q-recon-0829-farfields2` | ANSWERED | Do any of eight fields still absent from IMPORT-MAP.md and from the three 2026-08-28 recon notes hold a theorem whose hypothesis the tile or the zone satisfies exactly and whose conclusion is a lower bound on a periodic set's count in every window, a maximal-gap bound for a product-structured periodic set, or a positivity statement at a distinguished point? | No route from any of the eight; the pass banks one PUBLISHED-ANCHOR (Holt and Rudd's transfer matrix has its full spectrum in print, second eigenvalue a2 = prod_{17<=q<=p}(q-3)/(q-2), printed 0.10206751799779 at p = 999999999989, decaying only like 2.82/ln p, and this corpus's prior-art record names the eigenstructure and the binomial eigenvectors but carries the per-prime eigenvalue (p-j-1)/(p-2) at lit-pdf-holt-rudd.md line 259 but no product-over-primes closed form, no numerical value and no rate), one CLOSURE with mechanism (Beurling-Selberg is the proof device behind the large sieve, whose constant N-1+1/delta is an l2-to-l2 statement and whose error 1/delta exceeds the window at the operative theta_total > 2, so the whole extremal-function family reproduces a row already rejected), and three wall addresses; two angles are TPC-strength at the form that would pay and are labelled (ii). | 0 | [recon-0829-farfields2.md](history/staging/recon-0829-farfields2.md) |\n| `Q-recon-0830-rec-killrun` | ANSWERED | Does the literature hold, in its own conventions, a theorem whose hypotheses REC(s, u0) (the sup-versus-rms recovery of the level-D signed remainder over all positions, attack-0829n-rml-proof.md section 3) or the weighted kill-run K*(s) (the longest run of level-s slots the primes in (s, 2s] can kill, attack-0829n-doubling-bridge.md sections 0 and 3) satisfies as stated, or a Maier-type theorem that makes REC false for two-class sifted sets at some u0 below beta_2? | No theorem applies to either object as stated, on four calibrated channels searched in the owning conventions; every neighbour is graded NEAREST with its unmet hypothesis named, and neither object is on any refuted row. On the Maier question the answer is negative for REC as stated (the Maier family bounds the COUNT, i.e. the full-level remainder, never a level-D truncation) but carries one calibration: at one class and full level the REC-shaped inequality is FALSE asymptotically, by Buchstab's origin ratio against the Montgomery-Vaughan full-period variance ceiling, and the embedded arithmetic puts the level where that falsity first shows at log10 z between 16 and 141 depending on u0 and epsilon, so a finite-z margin of the kind the corpus measures cannot see a failure of this type; at two classes the same full-level statement is HL-conditional at u0 = 2 and unlocated in print at any u0. No exponent moved. | 0, D | [recon-0830-rec-killrun.md](history/staging/recon-0830-rec-killrun.md) |\n| `Q-recon-0830-smooth-aps` | PARTIAL | Does any theorem on y-friable integers in arithmetic progressions (Granville, Fouvry-Tenenbaum, Soundararajan, Harper, Drappeau, Drappeau-Granville-Shao) supply the uniform o(1) equidistribution that attack-0830-varE-identification.md section 4 leaves open, for moduli up to y^(4/5) with the lam1 weight and the squarefree restriction? | NONE APPLIES AS STATED, and the step stays open; lim Var/E = 0.45546 stays HEURISTIC. Read at the page (eleven sources, arXiv text layers and the Acta Math. page image): every pointwise asymptotic is O(log q / log y)-precise (Granville 1993 Thm 1) or hypothesises log x / log q -> infinity, and every beyond-square-root level (Fouvry-Tenenbaum 1996, Drappeau 2015, Drappeau-Granville-Shao 2017, Pascadi 2023/2025) hypothesises y <= x^delta, i.e. u -> infinity, against our u in (1.2, 2]. NEAREST: Harper arXiv:1208.5992 (2012, a preprint with no journal version found) Theorem 2, the Barban-Davenport-Halberstam form, whose range (log^K x <= y <= x, Q <= Psi/log^A x, bounded u in the ineffective form) covers d up to L^(1/2)/log and whose weight (1 on all friable n, no squarefree restriction, no max over x' <= x) is unmet. The record's quantifier is also refined: the cell needs the lam0-weighted AVERAGE over d, not uniformity. PROVEN given the weighted form (H_w) of that theorem, derivation mine and unreviewed, with its two identities checked numerically in the producer: every two-branch cell of Xmix, both mirror cases, all bands, balanced blocks included, is o(ln^2 y), so the Bettin-Chandee kernel separation is not on this route; the reduction of (H_w) to Harper's theorem is OUTLINED and not written, and the three-branch type (measured -0.0002 at x = 19, trivial bound O(ln^3 y)) is untouched. Not TPC-strength. | 9 | [recon-0830-smooth-aps.md](history/staging/recon-0830-smooth-aps.md) |\n| `Q-recon-0904-sifting-limit-floor` | PARTIAL | What is actually known, in the owning convention and at the page, about LOWER bounds on sifting limits at dimension kappa > 1, and is the class-blind cap at beta_2 = 4.26645 a barrier or a method artefact? | Both. The corpus's [ABSENT] is wrong for lower bounds and right for extremal examples: Selberg's Lectures section 17 and Brady's 2017 Corollaries 1 and 3 ARE lower bounds on beta(kappa) at kappa > 1, invisible because Selberg's convention is the reciprocal a_k = 1/beta_kappa so his lower bounds are titled \"upper bounds for sifting limits\", while no kappa > 1 extremal example exists and Halberstam's own 2003 review says so; the best floor on beta(2) resting on a citation is 1.8394 from Brady's reduction evaluated at kappa = 2, unconditional and below the band, and beta(2) >= 2 follows from monotonicity by a padding argument DERIVED here and probably folklore; the cap at 4.26645 as a METHOD ARTEFACT at rung MEASURED is REFUTED as stated (rider 2026-09-04, from `redteam-0904-sifting-limit.md`, on four independent grounds): the level-D LP fixes one omega-profile out of the class Ford defines by a one-sided inequality, and a legal profile inside the classical budget at x = 13 reads 5.0113 calibrated, above beta_2, firing the note's own pre-registered refutation; the ledger's \"3.3152 sits about 18 per cent below the DHR value\" is 22.3 per cent; and its \"passed, 0 of 15\" external test does not survive, so the calibrated sentence is that no barrier theorem exists, no axiom-only argument inside the band is exhibited, and the LP number is a finite-level measurement for one profile; the band (2, 4.26645] is untouched by anything proven. | 0 | [recon-0904-sifting-limit-floor.md](history/staging/recon-0904-sifting-limit-floor.md) |\n| `Q-reconcile-bind` | ANSWERED | Can the unbound script tails be reconciled against their code and then bound by embed.js? | Six producers came back CLEAN-BOUND and one REBOUND with two documented label supersessions and identical results; nothing was DEFERRED and nothing DRIFTED outside the one pre-registered band move, fdecay-band-01's point 35.5 to 33.9 and band [30.8, 42.9] to [29.6, 38.3]. | none | [reconcile-bind.md](history/staging/reconcile-bind.md) |\n| `Q-record-deficit` | MIXED (lit-kourbatov-shortfall.md: ANSWERED; record-location-null.md: PARTIAL) | Does the 6.0 percent record-location deficit survive the corrected null, and what is it? | Survives (z = -3.3, five window cuts); Z5's premises on 'location' and on n_eff were wrong; corrected-null formula in §4. | Z5 | [lit-kourbatov-shortfall.md](history/staging/lit-kourbatov-shortfall.md), [record-location-null.md](history/staging/record-location-null.md) |\n| `Q-record-mechanism-0830` | PARTIAL | Which of three candidate mechanisms (sub-Poisson gap dispersion, the tile's exact gap law, Kourbatov's k = 1 conspiracy heuristic transported to k = 2) carries the residual 0.78 to 0.97 of Kourbatov's b, at what derived size, and what fraction of b is left? | The residual is carried by the twin-prime gap law AT HEIGHT and by nothing else tried: a new sieve to 1e11 measures that law under-dispersed (CV^2 rising 0.7276 to 0.9295 over seven decades) with a far tail steeper than exponential (log-slope 1.0638 at the top decade), and fed into the record null it carries 95.7 to 99.8 percent of the residual (d b_z = 1.0370 and 1.0806 +- 0.0014 against 1.0833, MEASURED on a model, paired, 2000 replicates); the tile's period-wide law (d b_z = 8.10) and Kourbatov-Wolf's k = 1 bin-count heuristic (4.20) overshoot by 4 to 7x and are closed as posed; the transport of a bare CV^2 is family-dependent (0.62 against 1.06 at CV^2 = 0.93), so \"computable from Var/E\" is false as stated; Kourbatov-Wolf's second-order scale a_c/abar tracks the measured CV^2 to within 0.04 at every decade with no parameter and is an ansatz, not a derivation; the shape statistic moves a third of the way toward the data (z_D from -1.64 to -0.97); what is left with no derivation is the gap law itself, and the top height band's b (0.937 measured against 1.69 to 1.98 predicted on 17 records) is the open falsifier. | Z5 | [attack-0830-record-mechanism.md](history/staging/attack-0830-record-mechanism.md) |\n| `Q-record-null2` | ANSWERED | How much of Kourbatov's shortfall coefficient b does an inhomogeneous-intensity record null carry, how much does a 6Z-latticed null carry, and what is the residual? | The inhomogeneous null carries essentially none of b (d b_z = -0.0012 +- 0.0001, MEASURED); the 6Z lattice's share is LAW-DEPENDENT, +0.1113 +- 0.0003 under the memoryless rounding-up law against -0.0060 +- 0.0003 under rounding to nearest, so the null side is a bracket of 16.1 to 25.1 percent of b and three quarters or more stays a residual with no mechanism; record-location-null.md section 8's argued 1.7e-4 lattice bound is wrong but no single number replaces it; the 15 percent estimator disagreement is pinned as an exact weighting identity; and the shape assumption behind every b reading is measured at 1.6 sigma and stays open. | Z5 | [measure-record-null2-0829.md](history/staging/measure-record-null2-0829.md) |\n| `Q-records-placement` | ANSWERED | Do the seven unswept twin-gap records 76-82, above 2^53, land uniformly inside their stretches, or is there square-anchor coupling in record-start placement? | Sealed alone before the producer existed: the uniform band, the fixed definitions, the registered readings, the calibration abort gate and the riders are all fixed here and no placement fraction, q or stretch boundary was evaluated; scored in records-placement-01.md. | Z6 (retired) | [records-placement-01.md](history/staging/records-placement-01.md), [records-placement-02.md](history/staging/records-placement-02.md), [records-placement-prereg.md](history/staging/records-placement-prereg.md) |\n| `Q-redteam-0818` | ANSWERED | Do the six headlines of 2026-08-18 survive being checked by someone other than their author? | Four CONFIRMED, two WEAKENED, one REFUTED: the second instrument's independence claim is refuted, the growth law weakens to G2 = x ln^{2+o(1)} x with the o(1) positive, and B is polylog, Brady's Problem 3 and DP1's consumer stand. | none | [redteam-2026-08-18.md](history/staging/redteam-2026-08-18.md) |\n| `Q-redteam-0820-empirical` | ANSWERED | Do the two held empirical headlines, perfold-error-model and item-x-offset, survive an adversarial pass before integration? | Both SURVIVE: every decisive statistic reproduced digit-for-digit on independent code, the blind window was re-sieved and re-scored 33/37 with independently re-derived bands, and the cofactor purity claim is certified; two sentences are weakened (the anchor-direction identical is only bounded at 6-10 per cent, and the @31 sigma calibrated claim outruns its controls) and one sentence is refuted as stated. | none | [redteam-0820-empirical.md](history/staging/redteam-0820-empirical.md) |\n| `Q-redteam-0820-math` | ANSWERED | Do fekete-1d, smoothness-front and import-hypergraph survive an adversarial pass before integration? | Mostly CONFIRMED, with one REFUTED as stated (import-hypergraph's \"the engine can never carry a ln-power\" is false for bounded-size edges, though the row's verdict survives for FGKMT-sized edges) and one WEAKENED (the freeze survives, the frozen value does not, under the trusted ladder). | none | [redteam-0820-math.md](history/staging/redteam-0820-math.md) |\n| `Q-redteam-0820-night` | ANSWERED | Do the night's two staged headlines, mp-derivation.md and rho-maximal-law.md, survive an adversarial re-derivation before integration? | Both survive: every decisive number reproduced on independent code, the fresh window re-sieved with a different engine reproducing all 37 X, the prereg custody at commit 303711b confirmed with the anchor genuinely fresh, and all 37 sealed (mu, lam) values and all 74 band integers reproduced exactly. | none | [redteam-0820-night-empirical.md](history/staging/redteam-0820-night-empirical.md) |\n| `Q-redteam-0820-night-proofs` | ANSWERED | Do the two proof-heavy held headlines of the 0820 night, anchored-01 and hsub-01, survive independent re-derivation? | Both stand: zero refutations and zero corrections to any quoted number, with every digit re-derived from the staircase-note's own lemmas or recomputed from the definitions in independent code; the doubling inequality itself was not attempted and remains posed, open and trap-free. | none | [redteam-0820-night-proofs.md](history/staging/redteam-0820-night-proofs.md) |\n| `Q-redteam-0820-structural` | ANSWERED | Do the Mirror-Sweep Lemma and its palindrome, and the advmin@11 = 16 headline, survive an adversarial replay? | Both CONFIRMED on independent re-derivation with zero shared code, advmin@11 = 16 re-proven end to end so that 16 < 34 kills every class-uniform cap; the correction that matters is one quantifier made explicit in the inversion sentence, since a family whose conclusion reads the chosen classes is not capped by 16 at the anchored point, and two precision defects sit in a Thorne citation. | none | [redteam-0820-structural.md](history/staging/redteam-0820-structural.md) |\n| `Q-redteam-0821-exponent` | ANSWERED | Do attack-rhoms-01's exponent claim and attack-doubling-01's certificates survive independent re-derivation? | No numerical discrepancy anywhere in either target, every re-derived figure reproducing to the printed digit; the 3.5+o(1) cap survives the attack it was most likely to die to but is not an unconditional bound on G2, and two wordings that run ahead of their custody grades are WEAKENED. | none | [redteam-0821-exponent.md](history/staging/redteam-0821-exponent.md) |\n| `Q-redteam-0821-structure` | ANSWERED | Do the W1/HL identity and the anchored-02 ladder survive re-derivation from the definitions on independent code? | Both headlines survive with zero refutations and all ten sub-claims CONFIRMED, carrying two recorded corrections: the -0.33 se figure's sign is presentation-dependent, becoming +0.32 se when duplicate-theta folds are collapsed, and the shadow guarantee is stronger than sold, a one-line theorem rather than a 76/76 empirical. | none | [redteam-0821-structure.md](history/staging/redteam-0821-structure.md) |\n| `Q-redteam-0821-wave2` | ANSWERED | Do the three held analytic attacks, b2mean, f4weak and quartic, survive an adversarial re-derivation? | All three survive with every load-bearing number re-derived independently, including four full N_k curves on a different inclusion-exclusion implementation and a c1 column from a freshly fetched OEIS b-file; two sentences are weakened, the cert/C2 widening being non-monotone, and no claim is refuted. | none | [redteam-0821-wave2-analytic.md](history/staging/redteam-0821-wave2-analytic.md), [redteam-0821-wave2-structural.md](history/staging/redteam-0821-wave2-structural.md) |\n| `Q-redteam-0828-census` | ANSWERED | Do destroyer-census-01.md, stretch-01.md and records-placement-01.md survive an independent re-derivation of every load-bearing claim, and may they leave HELD? | All three survive on arithmetic, every decisive number reproducing exactly on an independent engine; one sealed reading (records READ-3) is REFUTED as vacuous and four sentences need correcting before integration. | Z0, Z4, Z6 (retired) | [redteam-0828-census.md](history/staging/redteam-0828-census.md) |\n| `Q-redteam-0828-closures` | ANSWERED | Do the five 2026-08-28 closure notes stand, and do they stand for the reasons they give? | Three stand on corrected sentences (1d, Z5b, 0); one stands in half (10, the 1.074 counter-claim does not); one does not stand for its stated reason (1c, closed on a narrower window than the prior work's significant one, and its CLOSED has overwritten a PARTIAL in the generated index). | 10 (retired), 1c (retired), 1d, Z5b (retired), 0 | [redteam-0828-closures.md](history/staging/redteam-0828-closures.md) |\n| `Q-redteam-0828-engine` | ANSWERED | Do the five 2026-08-28 certificate-engine notes survive an adversarial re-derivation, and does the one-class freshness factor correction hold before it reaches a live document? | The one-class factor holds and is stronger than either note states; four numbers are refuted (first s>=10.82 is x=263 not 239, first non-empty kappa=1 level is @37 not @53, rho*g_max is off by one gap, the @17 certified share and per-term constant carry the wrong convention) and one proof step needs the max-over-y form of BV; the cluster's shared verdict stands. | 11 (retired), 8 | [redteam-0828-engine.md](history/staging/redteam-0828-engine.md) |\n| `Q-redteam-0828-head` | ANSWERED | Do the three 2026-08-28 head-residual notes survive an adversarial re-derivation of every load-bearing claim on independent code? | Mostly yes on the numbers and no on two framings: \"h/R -> 1 iff beta CV^2 -> 2\" is REFUTED (the limit needs only boundedness, the product controls the RATE), and 9 to 59 percent of the quoted h - R is the choice of comparator, since R is the continuum functional and a discrete uniform origin sees R + 1/2 exactly. | Z4 | [redteam-0828-head.md](history/staging/redteam-0828-head.md) |\n| `Q-redteam-0828-litimports` | ANSWERED | Do the six notes of 2026-08-28 survive an adversarial pass, and which of their load-bearing claims are wrong? | Every measured number reproduces on independent code; Saffari-Vaughan Theorem 10's range exponent is 6/11 not 1/11, which voids import-fracparts's one banked THEOREM at fixed M and corrects row 15 twice; four further weakenings, one refuted verdict line, no note refuted whole. | Z5, 4 (retired) | [redteam-0828-litimports.md](history/staging/redteam-0828-litimports.md) |\n| `Q-redteam-0828-quadpoint` | ANSWERED | Do the four HELD quadpoint notes of 2026-08-22 survive an adversarial pass: the capture identity's proof line by line, the depth law and its asymptote, the sealed prereg scoring, and the prior-art citations at page level? | The identity is PROVEN once its window convention is written down and holds at every K of 50 anchors under an independent implementation; four load-bearing sentences are wrong (section 4's gloss on Sum capU, v1 section 2's limit, the prior-art note's 2%, and the note's unconditional y* sentences), and none of the four defects opens or closes a route. | Z0, Z2 | [redteam-0828-quadpoint.md](history/staging/redteam-0828-quadpoint.md) |\n| `Q-redteam-0828-varE` | ANSWERED | Does the 2026-08-28 chain from the exact comb variance to lim Var/E = Pr[GD(2) > 2] = 0.45546 survive an adversarial re-derivation, and does the refutation of the 0.611 reading hold? | The constant survives at HEURISTIC and the 0.611 refutation is CONFIRMED and strengthened (the frozen out-of-sample half of the protocol also fails on the control); the correction is that TWO steps are open, not one, and that the theta=1 branch is an exact identity with a read theorem rather than a second-hand numerical match. | 9 | [redteam-0828-varE.md](history/staging/redteam-0828-varE.md) |\n| `Q-redteam-0829-measure-a` | ANSWERED | Do the three held measurement notes of 2026-08-29 (g2z2, g2-generic, provenance/x37) survive an independent re-derivation of every load-bearing claim, a custody and pre-registration audit, and a label audit against the wrong-direction scheme? | 96 load-bearing claims checked, 88 CONFIRMED on independent code or by hand, 7 WEAKENED, 1 REFUTED; the one refutation is a label, not a result: the g2z2 note's M2 and its producer's printed block both call 4.177e-1 the summed Poisson probability of a two-pair zone over p >= 3, while the code sums every decade including p = 2, whose single contribution is 2.6586e-1, so the true p >= 3 figure is 1.518e-1 and the MISS verdict is unchanged; no pre-registration in any of the three shows any sign of a prediction edited after its run, and birth-time evidence puts the note file before the producer file in two of the three, with the generic producer's inode anomalous and its claimed timestamps uncorroborated. | none | [redteam-0829-measure-a.md](history/staging/redteam-0829-measure-a.md) |\n| `Q-redteam-0829-measure-b` | ANSWERED | Do the three 2026-08-29 HELD measurement notes (measure-roughpair-null-0829, measure-record-null2-0829, zone-tail-02-0829) survive independent re-derivation of their load-bearing claims, their custody and pre-registration integrity, their (i)/(ii)/(iii) labels, and their proposed edits to other documents? | 59 load-bearing claims checked, 44 CONFIRMED, 13 WEAKENED, 2 REFUTED, no data defect found and all three producers bit-honest under embed --check; the worst finding is custody clerical rather than numerical, measure-roughpair-null-0829.md:161 quoting code-sha256 4813584b against the producer's banner 5caaa5173b4c1561, a hand-transcribed hash that does not bind the artefact it names; the second is zone-tail-02's scope claim that every head bootstrap in the corpus is three times too narrow, REFUTED as written because destroyer-census-01.md publishes no bootstrap at all and head-residual-factor.js bootstraps over gaps rather than zones, while the defect itself is CONFIRMED on an independent engine (1,910 distinct a_first among 17,700 top-band zones reproduced exactly) and is LARGER than stated, a cluster bootstrap giving an inflation factor of 4.10 against the note's implied 3.04; measure-record-null2's exact cov identity reproduces to 1.4e-15 from the OEIS ladder and its lattice arithmetic reproduces at every row, but its headline \"wrong by about three orders in b units\" is unit-inconsistent and contradicts its own section 3e's \"about one order below the effect\"; no pre-registration shows any sign of post-run editing and none of the three can be shown NOT to have been, because all three preregs live inside note files whose mtimes necessarily postdate their producers'. | none | [redteam-0829-measure-b.md](history/staging/redteam-0829-measure-b.md) |\n| `Q-redteam-0829-measure-c` | ANSWERED | Do the load-bearing claims of measure-tail-deficit-0829.md, lit-scourfield-2008.md, recon-0829-farfields2.md and recon-0829-escapes.md survive independent re-derivation at the record, and are their custody, prereg, provenance and (i)/(ii)/(iii) labels intact? | No verdict of the four notes is overturned; of 61 load-bearing claims 38 CONFIRMED, 14 WEAKENED, 6 REFUTED, 3 UNVERIFIED. Worst: lit-scourfield-2008.md line 79 quotes lit-smooth-divisors.md section 6 as saying the convention has \"no asymptotic machinery in this range at all\", a string absent from the repository and contradicted by that section's own words, and uses it to justify a correction; measure-tail-deficit-0829.md section 2 describes a superseded embed and does not disclose the --force the header records, although embed.js --check passes on all three hashes; and note 1's reported M-P1 failure at 19 -> 23 is a single-draw artefact, reversed here on 40 draws (shuffled 296.1 sd 29.1, Markov 264.0 sd 18.4, share 34.9 percent inside the registered band). Note 2's precision correction inverts a quantifier, the open step needing o(1/ln y) against Scourfield's O(1/log x); note 3's corpus does hold the per-prime eigenvalue at lit-pdf-holt-rudd.md line 259; note 3's section 3c re-prices a family already resident as the Vaaler completion of its own Lemma V remainder. Every verdict, and most of the arithmetic, otherwise reproduces on an independently written engine. | none | [redteam-0829-measure-c.md](history/staging/redteam-0829-measure-c.md) |\n| `Q-redteam-0829-objects-gb` | ANSWERED | Does every load-bearing claim of the two 2026-08-29 object reads (G2 and the bridge) survive independent re-derivation at the cited record? | 60 load-bearing claims re-derived at the record: 47 CONFIRMED, 8 WEAKENED, 1 REFUTED, 4 left UNCHECKED and marked so; the refutation is the G2 note's L5 formula, which pairs the corrected trusted zone K in [1.3946, 11.3568) with the superseded single-base expression (ln 66 + K)/ln 16, an expression that evaluates to 5.607 above beta2 at the top of its own stated range, and the three live-layer edits already applied by the orchestrator are correct as written. | none | [redteam-0829-objects-gb.md](history/staging/redteam-0829-objects-gb.md) |\n| `Q-redteam-0829-objects-zm` | ANSWERED | Do the load-bearing claims of z2-state-draft-0829.md and object-models-read-0829.md survive independent re-derivation at the record, and is the Z2 draft fit to become research/Z2-STATE.md? | 96 load-bearing claims re-derived at the record, 84 CONFIRMED and 11 WEAKENED and 1 REFUTED; the one refutation is a custody label, since the Z2 draft grades as unstamped hand arithmetic a table that sits in an embedded producer's OUTPUT block, and the draft is fit to become Z2-STATE only after eight named line fixes. | none | [redteam-0829-objects-zm.md](history/staging/redteam-0829-objects-zm.md) |\n| `Q-redteam-0829-theorem1` | ANSWERED | Is Theorem 1 of attack-AB-bounded.md (B <= 9 A(z)^2 (E(z)-1) = O((log z)^8)) a theorem, and at what rung does the corpus carry it? | PROVEN, and the scope now covers the whole chain: the L1 to L5 mean-square derivation and Theorem 1's bound on B are each re-derived here with no hypothesis, so _H <= 9 A(z)^2 (E(z)-1) H holds for every z, s and H as a mean over one period, the record's [INFERRED] is an under-grade, and the live layer's \"published as Opera de Cribro 6.18\" is wrong on two counts and must be replaced. | none | [redteam-0829-theorem1.md](history/staging/redteam-0829-theorem1.md) |\n| `Q-redteam-0830-doubling` | ANSWERED | Do attack-0829n-doubling-bridge.md and attack-0830-doubling-killrun.md survive an adversarial re-derivation on an engine sharing nothing with their producers, and do REFUTED rows 94 and 98 stand as worded? | The mathematics survives: the maxsum certificate Ghat(2s) <= maxsum_{K*+1}(T_s) and the product composition K*+1 <= prod(1+L_j) are each re-derived here and hold, the second at 21,641,346 nesting links over 6,012,804 killed runs with zero failures including the zero-kill fold, and every quoted figure reproduces digit for digit on a fresh engine (K*(16) = 17; G2(31#) = 348 @ 8813641451 x4 from both base tiles; sup msc 6.6364; the diagonal cells; the 2^N ratios 1.226/1.481/41.348). Row 94 STANDS and is if anything under-claimed, but its attribution is wrong in one direction: Lemma 1 at s = 128 alone clears the whole band, so the s = 16 walk corroborates rather than carries it. One statement is falsified: the bridge note's NOT-REACHED line puts 19#->37# out of reach, but column-major it returns here, reproducing the ladder row x = 37 and adding a FIFTEENTH step at s = 19, 20 (K* = 13, N = 4, C2 3.5200, certificate 3.8000), so the enumerable range ends at s = 20, not s = 18. Row 98 STANDS on its mathematics and is WEAKENED on one clause: \"the truth sits under it everywhere (sup w/a = 0.8295)\" attaches the CERTIFICATE's ratio to the word truth; the truth's sup is 0.6591. | D | [redteam-0830-doubling.md](history/staging/redteam-0830-doubling.md) |\n| `Q-redteam-0830-engine` | ANSWERED | Do attack-0830-buchstab-deep.md, attack-0830-comb-tail.md and attack-0830-anchored-ladder-17.md survive an adversarial re-derivation on code that shares nothing with their producers, and does REFUTED row 97 stand as worded? | Every headline number in the three notes reproduces exactly on independent code (26 of 26 comparisons, including F2/f2 = 3.6682 at sigma = 4.7088 by a near-analytic route, the 0-of-1512930 pair count, the tail defect at five levels, the full 419328-term Legendre split, and the whole @23 anchored ladder including K* = 27, floorU = 5364, truth 597475 and the minimal pool 1543 proven exact); one load-bearing DERIVATION is REFUTED — \"\\|B - 1\\| >= c forces sigma(q) < u_c - 1\" fails at @23 on 21134 (q, K) pairs for the engine's model B at c = 0.1 and on 33875 high-count pairs for the measured B at c = 0.01, because the step bounds every u in a weighted average by the largest one and treats the denominator as e^-gamma, so the \"level-free\" clause of buchstab-deep section 4 and REFUTED row 97's \"\\|B - 1\\| >= 0.01 forces sigma < 1.798\" must be re-graded from DERIVED to MEASURED-at-@23 (where the conclusion survives: the worst sigma is 2.159, still far below beta2 = 4.26645); row 97's \"the best sigma any level has\" is separately WEAKENED, since @29's head sigma is 5.5785 with width 1.5325 and @97's is 17.1422 with width 1.0000; comb-tail derives its main-term factor from prod (1 - 1/q_i) and measures it against prod (1 - 1/(q_i - 1)), which is what natal-cap-28 line 283 computes against its own line 20 and against certificate-engine.md section 1, a 0.9% difference at @23 that moves nothing certified; comb-tail also misquotes its own cited source as F1 in [1.5, 4] where that source says [1.5, 3] and says there is no bound at all at s < 1; four smaller slips are listed and none opens or closes a route. | 8, A | [redteam-0830-engine.md](history/staging/redteam-0830-engine.md) |\n| `Q-redteam-0830-fekete` | ANSWERED | Do the load-bearing claims of attack-0829n-hsubpow-K.md (the sign lemma, the K -> beta_bound map, the zone, the TODO.md base-16 correction) and measure-0830-delta-reader.md (the six pre-registered readers, their kill gates, the joint fit's calibration, the RB circularity argument) survive independent re-derivation, and are the two live-doc corrections they caused the right sentences? | Every number in both notes reproduces on independently written code (all 14 cells of the beta_bound / K_min table, all 48 cells of the reader-calibration table, all 24 figures of the real-control table, the margins, the sups, the 5.607 correction, the 10-of-81 count), and the sign lemma is re-proven as stated for continuous exact laws. Two claims are WEAKENED and one live-doc sentence needs a qualifier: the sign lemma's constant 1.0597 loses 43% of its value if the base floor moves from 2 to 3 and its whole shape fails for the stepped law that a primorial ladder must present, where beta re-enters the sup and a delta = 0 law already needs K = 1.37 at beta = 2, inside the trap; and TODO.md 1d's \"the sign of delta is unreadable for G2 at reach 79 by any instrument tried\" is wrong on the letter, since reader RB passed all four pre-registered criteria and read delta_hat = +0.80 +- 0.09, the note voiding it post-hoc on a circularity the pre-registration flagged but did not make a kill criterion; the honest sentence adds \"non-circular\". No claim in either note is refuted. | 1d | [redteam-0830-fekete.md](history/staging/redteam-0830-fekete.md) |\n| `Q-redteam-0830-floor-growth` | ANSWERED | Does the DERIVED growth half of attack-0830-rec-cheapest.md sec.4.2 (Omega >> z^{16s/9}/ln^8 z from four-prime Rosser exit chains) survive an adversarial re-derivation on independent code, does it really make REC(s, u0) false for every u0 < 16s/9, and does the blind slope test of blind-0830-omega-floor.md test the LAW or the CONSTRUCTION? | No break found in six attacks, so the growth half SURVIVES this pass and stays DERIVED (one failed break is not a proof, and the whole thing still rests on the exact half E1-E3 that this pass did not grade); but four claims are corrected - the threshold 16s/9 needs the unstated hypothesis s <= 3 (above it the four-prime LP maximum is 2(2s+2)/3, then 8, both below 16s/9, though the corollary that every legal u0 < beta_2 fails survives at every admissible s because 2 x LP-max > beta_2 for all s > 9beta_2/16 = 2.3999), the 2s - o(1) limit needs 3^k <= ln D / ln 8 so only k = 2 is available at every z below 1.1e9, A_i is >= not ~ the dyadic count, and the certified ratio does NOT rise monotonically (0.0118, 0.0128, 0.0101, 0.0057, 0.0051, 0.0093, 0.0336, 0.0801, 0.1658, 0.3545 on my own recount, which reproduces all ten S4b rows digit for digit); the blind test is ruled a test of the CONSTRUCTION's count and not of the law, since it cannot see E1-E3 or the CRT step; on the ruling Chris asked for, the floor does make the RML route a truth gap asymptotically and conditionally on E1-E4, and changes nothing computable, the literal construction first passing H = z^{u0} at z ~ 5.6e31 on the constant 2.93e-4 measured here; no exponent moved. | 0 | [redteam-0830-floor-growth.md](history/staging/redteam-0830-floor-growth.md) |\n| `Q-redteam-0830-floor-sign` | ANSWERED | Does the EXACT half of attack-0830-rec-cheapest.md sec.4 (E1-E4, the sign facts, the window bound T <= cc(r) + H - 1, the exact Omega, and sec.(d)/(e)'s calibration sentences) survive an adversarial re-derivation on independent code, and does it by itself constrain REC at any u0 without the growth law? | The mathematics STANDS as PROVEN and is stronger in two places than the note claims, but it constrains REC at NO u0 by itself: E3's identity is pure algebra and holds at every position (0 mismatches in 20.5M positions, both s), E2 holds at every divisor n > 1 under three disjoint routes at every s >= 1 and FAILS at s = 0.8, locating a hypothesis D >= z the note omits, E4 has 0 violations at every x over four complete periods with the wrap, and Omega is reproduced by an independent enumeration at all 26 tabled values and EXTENDED to z = 101 (285, 441, 684, 990, 1155 at z = 79, 83, 89, 97, 101), confirming blind-0830's exact 684 and 1155; three sentences are wrong -- E1's \"every pair with gcd \\| 2 is realised\" (exactly half are; the missing half is all those with 2 dividing one side), the ledger's and sec.0's \"251 against z^{u0} = 1.05e6\" (1.05e6 is HM, z^{u0} = 7.19e7, so the gap is understated 68.3x), and sec.4.4's reason for lam+(P(z)) = 0 (0.24 exceeds 1/8; the figure that closes the slab is 0.1199 at z = 89, under 1/8 by 4 percent) -- and sup\\|rho~\\| >= (Omega - M)/2 is loose: sup\\|R_1\\| = Omega + M exactly. Every exactly known log_z Omega is under 2 while REC needs it above u0 > 2, so the exact half touches REC at no u0 and every adverse consequence in the note still rests entirely on the growth half. No exponent moves. | 0 | [redteam-0830-floor-sign.md](history/staging/redteam-0830-floor-sign.md) |\n| `Q-redteam-0830-imports` | ANSWERED | Do the six lighter notes of 2026-08-30 (varE-identification, smooth-aps, rec-killrun, skeleton-door, at43-bigint, coherence) survive an adversarial pass on independent code and at the source pages, and which of their load-bearing claims are wrong? | No note is refuted whole and no exponent moves. Every decisive measurement reproduces on code sharing nothing with the producers: the varE mixed remainder and its 0.844 / 0.826 share below 2L and 94.4 / 98.7 percent unbalanced share at x = 13, 17; the Buchstab deviations and all three crossing levels log10 z* = 33.53, 68.80, 33.62; the skeleton door's 9.2 / -0.9 percent, G30_agg 0.1113 / 0.1011 and the 0.0102 cap at @13, @17, by a path that also verifies cap-36 Theorem A; the @43 parity exactness on all 26,157,448 scour classes with 0 above 2^54 and a constructed @47 counterexample; and the five recon-0828 verdict counts summing to 75, which REFUTES rec-killrun's brief-error clause and confirms the orchestrator's rider. Seven claims WEAKEN. The load-bearing one is smooth-aps's \"the weight is unmet in print\": Harper's J. London Math. Soc. 112 (2025) e70293 proves a Barban-Davenport-Halberstam asymptotic for an ARBITRARY sequence on exactly that dyadic all-classes object, and his own footnote says the missing max over x' can be incorporated, so the NEAREST citation changes even though the step still does not close (the 2025 range is sqrt(2x) < Q <= x and its Theorem 2 route bars sieved sets by name). Also: Henriot Corollary 2 applies, but to the ERRATUM's statement and under two hypotheses the note leaves unstated (eps < alpha/600, not \"every eps\"; F at prime powers); varE section 4's display (*) carries an unfilled placeholder and is vacuous by a factor L/d; the skeleton door's \"would tighten\" is wrong in sign at @17, where the closable block is negative and removing it RAISES G30_agg to 0.1020; Halasz-Montgomery is not applicable for the reason given (it is an abstract inner-product inequality); FGKT's heuristic quote is on p. 4, not p. 3; and the \"~13x had no source\" is a corpus miss, the retired G2-walk figure at CHANGELOG.md:5716. Not TPC-strength; nothing here opens a route. | 9, 0, D, 1 | [redteam-0830-imports.md](history/staging/redteam-0830-imports.md) |\n| `Q-redteam-0830-records` | ANSWERED | Do attack-0829n-parity-dstar.md (CLOSED, HELD) and attack-0830-record-mechanism.md (PARTIAL, HELD) survive an adversarial pass on independent code, does REFUTED row 95 stand as worded, and is TODO Z5's new title supported? | Both notes' measurements reproduce on independent code and neither headline is refuted: D* is confirmed EXACTLY, by a BigInt rational simplex written here, at 299 of 376 anchors with 0 contradictions (the other 77 exceed the exact solver's size cap), the old range's 172-figure gate is re-derived exactly at all 43 anchors with 0 disagreements, and an independent 30-wheel sieve to 1e11 reproduces note (b)'s seven decade gap counts, CV^2 and far-tail slopes to every printed digit, with the counts also matching the published pi_2(10^k). Four sentences are wrong or over-stated and are replaced here: REFUTED row 95, note (a)'s ledger, its section 6 and TODO Z5 all attribute the null's 0.025 agreement to the WIDTH law when 0.025 is the ln Q slope difference (the width-slope difference on the same 84 matched anchors is 0.031); note (a) section 3's \"the threshold sits in a gap of at least 0.25 at every anchor\" is scoped in its own producer to the true arm and the null arm's exact gap is 0.1212; P6's HIT rests entirely on the decade-2 collapse that section 5 discounts as a sampler artefact when it discounts kill clause (a); and note (b)'s \"d b_z = 1.0806 +- 0.0014\" carries only the Monte Carlo pairing error, while moving the far-tail slope by 0.02 (against a spread of 0.0174 across six fit windows on the same decade and 0.1196 across decades) moves d b_z from 0.9888 to 1.1720, and continuing the law at slope 1 above the last resolved point still gives 0.7752. REFUTED row 95's CLOSED verdict stands on kill clause (b), which is better powered than the note presents it (paired s.e. 0.0401 against a 0.15 band), and the row needs two clauses reworded; TODO Z5's title is supported for \"measured and not derived\", and the resolved part of the law carries 72 per cent of the residual on its own with the extrapolation carrying the rest. | Z2, Z5 | [redteam-0830-records.md](history/staging/redteam-0830-records.md) |\n| `Q-redteam-0830-rml` | ANSWERED | Do the three HELD notes of 2026-08-29 night and 2026-08-30 on the RML chain survive an adversarial pass — the six-arrow chain and the term-count cap lemma of attack-0829n-rml-proof.md, the same-object ruling and the applied replacement sentences of verify-0830-usup-convention.md, and the reflection identity, custody, pilot bounds and sealed forecast of measure-0830-rho-sup-z41.md? | The three load-bearing PROVEN claims survive independent re-derivation — the CAP lemma (every step, sine sum at worst ratio 0.585568, the constant 6, 0 per-modulus violations at z = 13, 17), the reflection identity rho~(W-3-y) = -rho~(y) (proved here and its two hypotheses verified at z = 13..37, residual <= 1.7e-10 at every position of z = 13, 17, 19), and the same-object ruling (an independent lattice reproduces the repository's term multiset exactly at z = 13..37 and its {Vabs > 0} set equals LSE's records at all ten levels); five of the seven cited exact sups and three of the five walked sup\\|R_H3\\| reproduce to <= 5.0e-7, and every figure of the sealed z = 41 forecast reproduces from the seven cited sups alone. Fourteen claims do not survive, none of them reopening a route. The load-bearing one is RML sec.2's ms_m/ask column: it measures the PROVEN mean-square BOUND sqrt(B H), not the ^{1/2} that REC names, the two differing by 36.53x to 200.45x on the tabled rows because saturates in H while B H does not, so sec.5's ordering (ask below the Gaussian factor at u0 = 3.0 and 3.5) is an artifact and reverses at every row against the true mean square. \"gauss -> 0\" (sec.5) and \"sup/rms <= sqrt(2 lnW) O(1) = z^{o(1)}\" (sec.3) are the same error twice, against the note's own sec.0: sqrt(2 ln W) = z^{1/2+o(1)}, measured gauss 0.7905 at z = 19 falling to 0.5645 at z = 1e7, limit 1/2. \"the recovery exponent is log_z(2 sup\\|rho~\\|/rms): about 0.99-1.29\" quotes the truth column, which is log_z(2 sup\\|rho~\\|) alone; the quantity named measures 0.5628 to 0.6712. \"REC would be moot rather than false\" should read false. \"u_cap -> 2s + o(1)\" and \"n ~ z^{2s}/polylog\" are asserted where only \"<=\" is proven and the data cannot decide: u_cap is 4.2652 at z = 47, crosses beta_2 only at z = 53, reaches 4.4891 at z = 79, and passes u_triv from z = 53. \"seven points spanning less than one octave\" is 1.509 octaves, still uncorrected in three places including the ledger verdict. verify-0830's applied \"polynomially many from z = 223 on\" is true at every z >= 13, and its z = 331 crossover silently sets A = 1 (A fitted to LSE's own Ssat moves it to 337). No exponent moves, no route opens or closes, and no ledger verdict changes. | 0 | [redteam-0830-rml.md](history/staging/redteam-0830-rml.md) |\n| `Q-redteam-0830-slack` | ANSWERED | Do attack-0830-head-remainder.md, attack-0830-tail-derivation.md and verify-0830-record-defects.md survive an adversarial re-derivation on independent code, and do the five riders the verification put on live notes stand as written? | Arithmetically they survive: every figure of the three notes that this pass could recompute reproduced to the printed digit on code written from the definitions, 0 assertion failures over 58 assertion call sites (most inside per-level loops), including every figure of the verification's three claims and the half-decade Delta_HL it quoted rather than recomputed. Four sentences do not survive. (1) verify-0830-record-defects.md:316, a replacement queued for head-residual-hl3.md sec.0, says the pooled value sits \"0.038 to 0.052 below the sub-window mean at all three decades\"; measured, the two lower decades run to 0.076 and 0.066, and the same note's own falsifier row says 0.038 to 0.076, so the wrong half is the one queued to land. (2) verify-0830-record-defects.md:379's un-measured \"+0.005\" residual inside the eighths is +0.0004 measured at sixteenths, so the de-pooled top-decade value is converged and the caution can be dropped. (3) the ruling that the class null is \"the matched figure\" for the tail is matched on residue class only: p'^2 is coprime to every q <= p', and the ensemble's rough-class offset is E_rc - R = 5.6714 -> 7.5357 over x = 7..29 against the class null's 5.6000 -> 6.0336, giving t/(R_shell + 7.5357) = 1.0123 against t/classNull = 1.0157. (4) attack-0830-head-remainder.md sec.3's top row is not gap-scale matched: E[g] 244.0 against 235.9, y_match capped at 19997, and on the note's own six ensemble points Delta_ens ~ E[g]^-0.312, so meas/ens reads 0.9640 not 0.9539 and the headline \"0.954 to 0.995\" reads 0.964 to 0.995. Three levels the notes recorded as out of reach are run here: the X2 group at x = 23, the tile identities and conditionals at x = 29, and the ensemble ladder at y = 29 (pipe/ens 0.9535, 40 s, against the head note's \"about fifteen minutes\"). No route opens or closes and nothing here touches Z2. | Z4, 9 | [redteam-0830-slack.md](history/staging/redteam-0830-slack.md) |\n| `Q-redteam-0830-zone` | ANSWERED | Do attack-0829n-X-upper.md, blind-0830-quadpoint-31607.md and blind-0830-quadpoint-c1c2.md survive an adversarial re-derivation and an independent re-implementation of their producers, and at what rung does the corpus carry each of their headline claims? | They survive on the numbers and mostly on the wording. On independent code every band sum, tier ratio, sealed forecast, band statistic and z of all three notes reproduces to the printed digit (X-upper 7.82 to 10.13 and the flat 378100.0; the fourth decade's 76 rows at 64 HIT and 12 MISS with the same twelve MISSes; the fifth decade's class A at both bands), and three attacks failed to break them (the paired-se attack on the D rows, a search for a fitted constant in C1, and gap-selection at the band edges). Four defects stand: the X-upper note's \"s_max 1.81 to 1.96\" is wrong in its own arrow convention, which means B3 to B8, since s_max peaks at B4 and reads 1.85 at B8; the c1c2 note's OLD R2 row at B14 is a boundary call that my 1-in-8 subsample scores the other way; the c1c2 note's custody line \"three tails, one code-sha256\" is a tautology of tailfmt.js and is not evidence of a seal; and the unscored y* check carries a Jensen bias of about +0.44 in y, which is the whole size of the C2 gap it is read as corroborating. Nothing here moves 4.2665 to 2, no rung rises, and all three notes stay HELD. | Z2 | [redteam-0830-zone.md](history/staging/redteam-0830-zone.md) |\n| `Q-redteam-0904-argmax` | ANSWERED | Does measure-0904-argmax.md survive an adversarial re-derivation of its positions, its congruence null, its residue chi-square and its own pre-registration scoring? | Every positional number in the note reproduces exactly on an independent fold-replication engine at x = 11..31, including the argmax lists, m(x), the top-K sizes and components, the per-prime pair counts, the tail counts and the F6 drops, so the note's data survive. What does not survive is three things and the largest is the null: two random TWIN SLOTS are congruent modulo W/p with probability (p-3)/(D-1), not 1/M, so the note's \"10^6 to 10^8 times the uniform null\" is inflated by 6.9 to 29.1 at these levels and should read 10^4 to 10^6, and the statistic is not a random variable at all but a deterministic function of each window (translate lemma, proven here and checked against every argmax pair count), which makes a null ratio a category error rather than a small correction; the note's own \"58 to 82 per cent are translates of another\" reports K - components when its wording needs K - isolated, which is 83 to 98 per cent; and the custody table in §2 quotes a code-sha256 (3019147f...) that the producer does not carry (d53a3fd1...), while §2's body and §5(5) still describe a widths annotation the rider says was already deleted. The scoring is faithful in verdict and wrong in arithmetic: F3 and F7 are scored at four levels, not five, so F7's falsifier fired at three scored levels and F3's tally is 2 of 4; the brief's \"three of eight\" does not exist on disk, both the ledger and §0 say four, and four is right. The exponent is untouched and nothing here is a route. | none | [redteam-0904-argmax.md](history/staging/redteam-0904-argmax.md) |\n| `Q-redteam-0904-floor-growth-2` | ANSWERED | Does the growth half of the adverse pointwise floor (Omega >> z^{16s/9}/ln^8 z from four-prime Rosser exit chains) survive a SECOND adversarial pass, and does a floor on the certificate at a CRT-planted window make REC(s,u0) false (a truth gap) or only unprovable by that certificate (a proof gap)? | SURVIVES, at the same rung and no higher: four new attacks and two unasked ones found no break, so the growth half stays DERIVED and red-teamed twice, which is two failed breaks on a derivation and not a proof; the proof-gap reading is REFUTED because REC is a statement about R_H, the remainder of the very certificate the floor bounds below, and the chain has no step where a certificate value must become a true remainder, so it is a truth gap for REC as the corpus states it; five corrections are owed to the record -- the REFUTED row is UNSCOPED and should read \"on the Rosser vector-sieve lattice at equal levels\" since only the Rosser instance is derived, the \"falsity beyond z ~ 5.6e31\" is an s = 2.698721 figure quoted without its s and the fitted crossing at the corpus's own measured family s = 3.0 is z ~ 10^15.3, the analytic p* window (4D^{1/27}, D^{1/9}/4) is neither necessary nor sufficient at reachable levels so the exact onset is 4.42e5 at s = 3.0 rather than the quoted 1.055e6, the four-prime construction has NO exit prime below z once s >= 7 so the first pass's \"8 beyond s = 5\" holds only on [5,7), and the first pass's escape-(a) density 1/W understates the planted class by the factor W/(n1n2) though the escape still fails by an exponential; the LP maximum 8s/9 is confirmed as an exact vertex and the corollary crossing reproduces 9beta2/16 to 4.4e-16, the pre-registered slope band held at a new s (top step 4.9025 at s = 3.0 against a model 4.8990), and no exponent moved. | 0 | [redteam-0904-floor-growth-2.md](history/staging/redteam-0904-floor-growth-2.md) |\n| `Q-redteam-0904-item0` | ANSWERED | Do the two HELD notes of 2026-09-04 on item 0, the second adversarial pass on the adverse floor's growth half and the L7 one-class-to-two-class transfer, survive an independent adversarial pass, and do the live-layer changes they propose survive verbatim? | Both notes SURVIVE at their own rungs and neither moves an exponent, but SIX of their proposed live-layer changes need rewording and two of their supporting derivations are wrong; every load-bearing number reproduces on independent code (18 of 18 A1A2 cells digit for digit, the LP maximum 8s/9 at 0 deviations of 16 with residual 4.441e-16, the crossing 2.399878284834 against 9beta2/16 at 1.097e-13, the fitted C = 1.1206e-4 and the s = 3.0 crossing 10^15.30, the onsets 4.390e5 / 1.111e6 / 1.860e6, ln(n1n2) = 7.4993e7, and the L7 CRT counts EQUAL at x = 7, 11, 13), so the truth-gap reading of REC is CONFIRMED and the growth half stays DERIVED and red-teamed twice, not proven; what does not survive is the scoping, the exit-window and the price wording: the proposed clause \"at equal levels D1 = D2 = z^s\" is NARROWER than the derivation supports, since over 1643 admissible upper-and-lower level splits the floor beats the decoupled window at every cell by at least 0.644444, so no level split escapes and the sibling rider \"asymmetric levels give 8s1/9 + 8s2/9 with s1 + s2 pinned\" is REFUTED as written (it holds only while both s_i <= 3); the record's z ~ 5.6e31 is u0-specific as well as s-specific (10^31.68 at u0 = 4.2165 against 10^35.36 at u0 = beta2, same s), so replacing it with a beta2-convention 10^15.30 mixes two conventions; the onset correction is right numerically and wrong in wording, since at the record's own s = 2.698721 the exact onset is 1.875e6, LATER than the quoted analytic 1.055e6, not earlier; the escape-(a) correction is right in substance but its \"(A1A2)^2 / (2 n1n2)\" is a LOWER bound on the split's contribution, and the two-sided version through lambda+(n) <= 2^omega(n) reads 10^{-3.004323e+7} against B ~ 10^{10.1}, still no contradiction; and on the L7 side the union-bound argument and the CRT count are CONFIRMED while the coupled vector-sieve price 2(1 + sqrt e) = 5.297442541400 is the FORCED-equal-level constant, the corpus's live figure being K_BF = 5.158064680330, reproduced here to twelve digits at (a, b) = (2.229949, 2.928115), and the sieve-on-holes is the vector sieve only after the Ford-Halberstam substitution step that sift-limit-attack.md 7b(1b) already owns; the \"no third mechanism excluded\" line is honest but incomplete, since weighted sieves with Chen switching are already priced and closed at sift-limit-attack.md 4.2; scope of death gains one row, L7's mechanism 2, which the growth note's own section 10 calls untouched while its sibling calls it dead; no exponent moved. | 0 | [redteam-0904-item0.md](history/staging/redteam-0904-item0.md) |\n| `Q-redteam-0904-r0-extension` | ANSWERED | Do the seven load-bearing claims of derive-0904-r0-extension.md survive an adversarial re-derivation against Tao's own text, and do its proposed live-layer changes survive? | Six of seven survive at their stated rung and one is refuted: Proposition A, the scope slip, row e'' and the equivariance transfer are CONFIRMED (the slip is strengthened, since Tao's own Example 2 is the target note's missing one-coordinate witness), Propositions B and D are WEAKENED (B's stated mechanism proves too much and its proposed insertion drops the qualifier that saves it; D's \"same trigger\" is near-definitional and its causal reading of the exponent band does not hold), and the claim that no corpus artifact certifies the window-count variance at width p'^2 is REFUTED by research/06-variance-theorem.js, which computes exactly that, brute-force verified at two levels; separately, EXEMPT throughout means \"Claim 1's H5 fails\", which is not the parity statement wall-note Face 1 is priced against, and the note's proposed live-layer text carries no H6 conditional. | Z2 | [redteam-0904-r0-extension.md](history/staging/redteam-0904-r0-extension.md) |\n| `Q-redteam-0904-sifting-limit` | ANSWERED | Do recon-0904-sifting-limit-floor.md's five claims survive adversarial re-derivation at the page, and in particular is the 4.26645 cap a method artefact at rung MEASURED? | Four of five claims survive with corrections and the headline does not. Claim d is REFUTED as stated on four grounds, each sufficient: the level-D LP fixes one omega-profile out of a class Ford defines by a ONE-SIDED inequality, and inside the classical (Omega_2(kappa,L)) budget a legal profile at x = 13 gives a calibrated reading of 5.0113, above beta_2 = 4.26645, which fires the target note's own pre-registered refutation of its floor; the calibration is a 1.72 multiplicative correction fitted at one point and it misses the second proven point beta(1/2) = 1 by 66 to 82 per cent, two and a half times the 28.7 per cent effect the headline rests on; the pooled 3.3152 averages four statistics of which only s* is the definitional analogue of beta, and that route reads 3.4678 here and 3.9487 at x = 43 in the corpus's own data, still climbing at +0.4297 per ln x with the fitted line reaching 4.26645 at x = 104; and 3.3152 is 22.2961 per cent below 4.26645 rather than the 18 per cent the note prints. Claim a is CONFIRMED with a second source off the OCR channel (Ford 2023 p. 37, \"Some authors, e.g. Selberg, refer to 1/beta(kappa) as the sieve limit\"), one page correction owed (Halberstam is p. 117, not p. 116). Claim b is CONFIRMED and strengthened: Brady's class is stated on his p. 10 and IS the axiom-only interval class Face 4 needs, his v_R table is reproduced here, and the identity v_{2d+2} = v_{2d+1} his proof asserts is confirmed at d = 0, 1, 2, so 1.819592 rests on proved facts alone. Claim c is CONFIRMED but demoted: beta(2) >= beta(1) = 2 is two lines from Ford's own one-sided (Omega) plus his own beta(1) >= 2, so the padding construction is correct and unnecessary. Of the eight proposed live-layer changes, two survive verbatim, five need rewording and one is refuted; the new SEARCH-CONVENTIONS row adds exactly three clearing phrases and changes nothing for any existing absence claim. | 0 | [redteam-0904-sifting-limit.md](history/staging/redteam-0904-sifting-limit.md) |\n| `Q-redteam-at37-score` | ANSWERED | Does the @37 scoring record survive an adversarial pass on its prereg re-score, its independence claim, its interpretation table and the cofactor extrapolation? | The empty-survivor scoring holds - every registered figure reproduces from the committed prereg text and the embedded census block, and the four skipped reference-level gates were run here and all pass - with two findings against the held report, neither touching the verdict: the one-level-drop figure is 8.2x rather than 4x, which strengthens the reversal claim, and the cofactor-one surplus sentence oversells. | none | [redteam-at37-score.md](history/staging/redteam-at37-score.md) |\n| `Q-redteam-zonegap-foundation` | ANSWERED | Do natal-onset-01, zonegap-01 and zonegap-prior-art, the three artifacts the Z2 pivot builds on, survive an adversarial pass? | The foundation HOLDS: all four natal-onset claims CONFIRMED, zonegap-01 confirmed at every reachable point by independent code using a different sieve and a different max-gap algorithm, and the prior-art spot checks all confirmed at page image; two prose defects found, both WEAKENED and neither refuted. | none | [redteam-zonegap-foundation.md](history/staging/redteam-zonegap-foundation.md) |\n| `Q-refuted-audit-0829-1` | ANSWERED | Do the closures recorded in REFUTED.md rows 1 to 17 survive re-derivation against the object definitions, the producers' embedded output and the corpus corrections through 2026-08-29? | Ten of the seventeen rows stand as written and seven are SOUND-NARROWER on wording that claims more than the record carries; no row is WEAKENED or UNSOUND, no route reopens, and the sharpest defect is row 16 crediting a published lemma whose own hypothesis excludes this corpus's operative range. | none | [refuted-audit-0829-1.md](history/staging/refuted-audit-0829-1.md) |\n| `Q-refuted-audit-0829-2` | ANSWERED | Do the closures recorded in REFUTED.md rows 17 to 34 survive re-derivation against the object definitions, the producers' embedded output, the 2026-08-29 corrections, and their own quantifiers? | All eighteen closures stand and none reopens; nine rows are SOUND and nine are SOUND-NARROWER, the worst being row 53's \"the search is finished\", which its own record contradicts in two places, and the others splitting into three classes: a mechanism clause that covers half its route (rows 47, 58), a universal quantifier wider than the sweep behind it (rows 49, 59), and a measured number carried without its calibration or its unit (rows 43, 46, 60). | none | [refuted-audit-0829-2.md](history/staging/refuted-audit-0829-2.md) |\n| `Q-refuted-audit-0829-3` | ANSWERED | Do the closure arguments behind REFUTED.md rows 34 to 51 (the greedy oracle through the bounded-differences family) hold when re-derived against the object definitions, the producers' embedded OUTPUT blocks, and the corpus corrections landed through 2026-08-29? | The band holds: 13 of 18 rows SOUND and 5 SOUND-NARROWER, no WEAKENED and no UNSOUND, so no route reopens; the worst finding is a factual inversion in the anchored-delta row, which asserts that delta exceeds the forced scale at 2 of 7 levels when the record's own table has \\|delta\\| < F at 7 of 7, and the same wording has already propagated into a second note. | none | [refuted-audit-0829-3.md](history/staging/refuted-audit-0829-3.md) |\n| `Q-refuted-audit-0829-4` | ANSWERED | Do the closures recorded in REFUTED.md rows 51 to 67 hold when the record is opened and the closing mechanism re-derived against today's corrected corpus? | All seventeen closures stand: 8 SOUND and 9 SOUND-NARROWER, zero WEAKENED and zero UNSOUND, no route reopens; the narrowings are wording that claims more than the record carries, worst of them row 67, whose closing clause quotes a section its own record headed POST HOC and carrying no verdict after the pre-registered primary went VOID on its matched control. | none | [refuted-audit-0829-4.md](history/staging/refuted-audit-0829-4.md) |\n| `Q-refuted-index` | SUPERSEDED | Where do recorded citations to the closed-route index resolve? | The current closed-route records are in OUTCOMES.md. This file is only a compatibility pointer for citations in recorded script output; it contains no separate outcome register. | none | [REFUTED.md](REFUTED.md) |\n| `Q-registry-glossary` | ANSWERED | What does each word of the programme mean, one term per object? | The vocabulary registry; adopted terms with dates, the anchored layer, and the calibrated entries (Var/E, hyperuniformity, the freshness dimension count) as corrected on 2026-08-28. | none | [GLOSSARY.md](GLOSSARY.md) |\n| `Q-registry-outcomes` | ANSWERED | What has each research approach established, where has it failed or stopped, and what evidence or changed input permits reuse or a further attempt? | One outcome register organized by research question and approach. Each record preserves its calibration, established result, limitation or failed step, evidence and reuse or revisit conditions. The closed-route table is part of this register. Git owns revision and authorship history; an open estimate is not a refuted claim. | none | [OUTCOMES.md](OUTCOMES.md) |\n| `Q-registry-research-readme` | ANSWERED | Which file in research/ answers which question, and in what order should a reader open them? | Router only: state lives in ../README.md section Status and G2-STATE.md section 0, the target in ZONE-POSTULATE.md, the queue in ../TODO.md; this file holds no mathematics and no summary of the programme. | none | [README.md](README.md) |\n| `Q-research-execution` | ANSWERED | How should agents use the integrated returns, avoid repeated work and select the next bounded research obligation? | The second dispatch from 1285d47 closed at 46fa948 and was independently reviewed on 2026-09-09. A/A2/B/C/D/E and V/W returns are present and integrated at their owning scopes. A2's corrected Type I estimate is accepted; D's fourth residual cut is retained and its remaining moment saving corrected to 7/200. C's low nonzero modes and E's aggregate scope are corrected. No new assignment is started by this integration; next obligations are candidates, not running work. Twin-prime infinitude and every sufficient signed margin remain OPEN. | C | [RESEARCH-EXECUTION.md](RESEARCH-EXECUTION.md) |\n| `Q-research-handoff` | ANSWERED | What must an incoming research agent understand, verify and return before its work can advance the current twin-prime argument? | Current handoff after the 2026-09-09 independent review of the dispatch ending at 46fa948. The full residual has four simultaneous complement conditions after D's regional estimate. A's centered truncation and A2's repaired low Type I estimate give the alternative S=C2*x+B+O_A(x/log^A x). B and all sufficient signed margins remain OPEN. The C3 absolute bound, family identities and scoped C/E failures retain their owning hypotheses. Execution is owned by RESEARCH-EXECUTION.md. | C | [RESEARCH-HANDOFF.md](RESEARCH-HANDOFF.md) |\n| `Q-research-round-validation` | ANSWERED | Which submitted results from lanes A, A2, B, C, D, E, V and W survive direct review, and what can be integrated or used to choose the next attempt? | Reviewed incoming commits 07ab47f through 46fa948 on 2026-09-09. A's truncation and A2's repaired low Type I estimate are accepted with bookkeeping corrections. B's ordinary-BV family identity survives; its uniform Dickman constant and complement condition are repaired. C's norm obstruction survives on all smooth inputs, but low nonzero Fourier modes are also 1-bounded. D's block estimate and fourth residual cut survive; its next moment saving is 7/200, not 17/200. E's rough-product distribution and aggregate constant 4 are accepted at the stated Wu uniformity reading; rational certificates close the two displayed tests for every fixed u>4. W remains OCR-only at the original 1974 theorem page. All sufficient twin-prime margins remain OPEN. | C | [research-round-validation.md](research-round-validation.md) |\n| `Q-residual-coverage` | ANSWERED | What portion of the full residual is controlled with uniform margins, and can the exceptional first branch improve that coverage rather than just a regional corner? | The first-branch and prime-free moments remove the exceptional norm condition. Full rectangles are controlled when delta+nu<19/25 or 5delta+2nu<123/50. Uniform twists justify exact coupled cuts: every fixed de<=x^theta with theta<19/25, and d^5*e^2<=x^lambda with lambda<123/50, as well as their union, have arbitrary fixed logarithmic savings. A concrete remaining endpoint sum has de>floor(x^(3/4)) and d^5*e^2>floor(x^(49/20)). Its twin margin is OPEN. The right zero-frequency budget, after an extra Cauchy inequality, limits the uniform product range; the left cross moment limits the other edge. Finite checks validate identities and masks, not asymptotic rates. | C | [residual-coverage.md](residual-coverage.md) |\n| `Q-review-0905` | ANSWERED | Does the present programme justify its latest stopping decision, and what is the next useful campaign toward infinitely many twin primes? | The computational checks pass and the DHR application survives this reading, but several summaries overstate the obstruction or misstate an implication. Recommend repairing those claims, retaining the fold framework, and testing it against Chen on long intervals before attempting a specifically stated parity-sensitive estimate. No proof of twin primes or improvement of an exponent is supplied. | none | [review-0905.md](history/staging/review-0905.md) |\n| `Q-review-request-0906` | ANSWERED | Which claims of the arithmetic campaign should an independent reviewer check, in what order, against which records, and what must the review return? | Completed review specification at the bounded scope recorded in handoff-review-0906.md and reports 20 and 21, with the subsequent round reviewed in round-review-0906.md. The dependency list records questions at dispatch, not current unfinished assignments. Imported proofs retain their review limits and the twin margin remains OPEN. Current assignments are in RESEARCH-EXECUTION.md. | C | [REVIEW-REQUEST.md](REVIEW-REQUEST.md) |\n| `Q-review-request-0907` | ANSWERED | Which claims of the 2026-09-07 session should an independent math agent check, in what order, against which records, and what must the review return? | Reviewed 2026-09-08 by an outside reader (history/reviews-0907/10): ranked claims 1 (conditional on the unread 1974 page), 2, 4, 5, 6 verified within stated scope; claim 3 verified with its quoted tolerances corrected (max deviation 1.8e-4, not 1e-4 or 1e-5); the Corollary 1 slip sharpened (fails for every representative other than 1, and mis-signs even then); ASSUMED items 4 and 6 discharged, 2 partially, 1, 3, 5 not. Twin-prime infinitude and every sufficient signed margin remain OPEN. Ranked below: the §6.2 demotion in the two-class manuscript (a change to a proof's dependency), the D_y identity reading (a change to what a census can show), the Erdős #687 attribution, and the two measurement artifacts. Twin-prime infinitude and every sufficient signed margin remain OPEN. | Z0 | [review-request-0907.md](review-request-0907.md) |\n| `Q-rho-maximal-law` | PARTIAL | What is the rho maximal law, what is its weakest sufficient form against the crossing, and where does it sit against print? | The pricing lemma turns any RML(alpha) with alpha < beta_2 = 4.26645 into G2(z#) << z^alpha ln^2 z, every step proven except the law itself; the sufficiency curve lambda_max falls from 2.31949 at z = 13 to 1.78446 at z = 47, the weakest measured point, and whether it ever reaches zero is model-dependent and undecided, so nothing here is an unconditional exponent. | 0 | [rho-exact-z31-01.md](history/staging/rho-exact-z31-01.md), [rho-maximal-law.md](history/staging/rho-maximal-law.md) |\n| `Q-rho-sup-z41-0830` | PARTIAL | What is the exact sup over all positions of the centred sawtooth remainder at z = 41 (period 7.42e12), and of R_H at H = z^3 and z^4, and what does the eighth exact point do to the truth's slope and to the u0 = 3 form of REC(s, u0)? | Not run in full; the exact z = 41 point prices at 4.8 to 5.4 h on this machine from measured per-position costs (naive 69.8, table 23.5, table-plus-windows 36.6 ns per worker; fleet 4.52 ns of wall at z = 41), above the 3.5 h rule, so what exists is a new engine (a PROVEN reflection that halves the walk, a P(29) periodic table worth 2.9x, per-block exact re-seeding) reproducing every cited exact sup at z = 13..37 digit for digit, sup\\|R_H\\| at H = z^3 walked for the first time at z = 19..37 (F3 = 26.4 to 35.9, the factor 2 was 1.5 to 2.2x loose), a sealed pre-registration (forecast sup 60.07, band [47.25, 76.37], kill needs sup\\|R_H3\\| >= 1807.417), one pilot twelfth giving lower bounds sup\\|rho~\\|(41) >= 70.651250 and sup\\|R_H3\\| >= 64.416936, and an eleven-segment plan for the box; no bound-and-prune exists on three routes; nothing asymptotic can move whatever the point turns out to be. | 0 | [measure-0830-rho-sup-z41.md](history/staging/measure-0830-rho-sup-z41.md) |\n| `Q-rho2-bound` | ANSWERED | Is there an analytic upper bound on (z) from its proven closed form, and does it matter for RML? | Already proven 2026-08-21 (attack-rhoms-01); the constant is now named, 170.88 z^{2s} ln^8 z; Chebyshev off any second moment diverges from RML (the true C_L = ms = 1 floor clears at z = 41 and 43 and misses by 2.155x at z = 47), so it is an ingredient, not a route. | 0 | [attack-rhoms-01.md](history/staging/attack-rhoms-01.md), [rho2-analytic-bound.md](history/staging/rho2-analytic-bound.md) |\n| `Q-rml-proof-0829n` | PARTIAL | Starting from Lemma V's PROVEN mean-square bound, what is the exact chain of implications to a two-class exponent below beta_2 = 4.26645, which single arrow is open, and does it close? | The chain from the proven mean square to an exponent below beta_2 has exactly one open arrow, the mean-square-to-sup recovery REC(s,u0): sup_x \\|R_H(x)\\| <= z^{u0/2 - eps} rms(R_H) at H = z^{u0} for one fixed u0 in (2, beta_2); every attempt here dies at a named place (term-by-term accounting is capped at 2s + o(1), the trivial-level exponent; position-union pays e^{theta(z)/q}; window smoothing and the P(y) lever move polylog only; the smooth-profile form is closed on hypothesis and priced above beta_2), the arrow is measured TRUE with certified margin z^1.72 to z^1.92 at u0 = 4 over z = 19..37 and the truth's slope 2.766 +/- 0.212 sits 5.8 se under u0 = 4 on seven points under one octave, so the failure is a proof gap at every computable z and undecided asymptotically; no exponent moved. | 0 | [attack-0829n-rml-proof.md](history/staging/attack-0829n-rml-proof.md) |\n| `Q-roughpair-error` | PARTIAL | How big is the empirical error of the rough-pair census, and is Z2 dead on the numbers? | Not dead on the numbers - the error sits below T at every one of 1,206 anchors, its scatter is sub-Poisson, its systematic part is 0.1% at the top band with the favourable sign and its exponent is 0.40 below the twin count's - but the same run moves the wall off the error term onto the depth, where the slack at the crossing is 1.056 and falling and the main term alone is 5.6 times the twin count at the sifting limit. | Z2 | [attack-roughpair-error.md](history/staging/attack-roughpair-error.md) |\n| `Q-roughpair-null` | ANSWERED | Does the independent-thinning null named in attack-roughpair-error.md's defects list return chi2/df = 1, and is the measured sub-Poisson dispersion of the rough-pair census error real or a normalisation artefact? | MEASURED, null-side, label (i), no bearing on the certificate: the independent-thinning null is binomial and returns 1 - p (0.8164 to 0.9808), not 1, so attack-roughpair-error.md's \"below 1 everywhere: sub-Poisson\" is WEAKENED and the registered F1 fires at u = 5; the deficit against the mean-matched null survives at pooled z = -5.87, -7.09, -11.39; and a CRT-exact null that only makes each prime's kill count exact in the window predicts 0.33 to 0.56, which the measured value EXCEEDS by 1.26 to 1.40, so D8 is interval-level in form, elementary in content, and not wall-relevant. | Z2 | [measure-roughpair-null-0829.md](history/staging/measure-roughpair-null-0829.md) |\n| `Q-round-review-0906` | ANSWERED | Which conclusions of the completed agent round survive review, and which corrections change the next research specification? | The fixed-band lift and continuous moving-window estimate survive the checked source interface. Interval stability additionally gives a dyadic scale-average logarithmic saving, so the blanket scale exclusion was too strong. The rate and missing prime-cofactor terms with s>1 or s'>1 still prevent a twin margin. Proper-prime-power branches already have a negligible bound on the fixed corner. The driver computes integer parity, not the CRT gcd partition; its total coefficient identities survive. Diagonal lower-bound, reachability and six-cut bookkeeping overclaims are corrected. No region, exact residual cut or twin lower bound changes. | C | [round-review-0906.md](round-review-0906.md) |\n| `Q-row11-closure` | CLOSED | Can the (1 - 1/e) greedy-coverage certificate of IMPORT-MAP row 11 ever fire? | No, and now on a producer rather than on prose: the map's paragraph-only pre-pricing reproduces exactly at all six figures it states and extends to the full ladder x = 5..79 with the certificate silent at every level under every constant that could be used, and the step from a density to an algorithm is proved in one line. | none | [row11-closure.md](history/staging/row11-closure.md) |\n| `Q-row11-closure-prereg` | OPEN | Does IMPORT-MAP row 11's (1 - 1/e) certificate reproduce, and does its test ever fire? | Pre-registration only, hand-written and committed before its producer existed: the claim being reproduced, what the producer must return, the pass criteria and the disclosed leaks are all fixed here, with the killing arithmetic still living only in prose. | none | [row11-closure-prereg.md](history/staging/row11-closure-prereg.md) |\n| `Q-row12-lonely-runner` | CLOSED | Does the lonely runner conjecture, or the Birkhoff and bounded-remainder branch, transfer to the Zone Postulate or to G2? | CLOSED WITH MECHANISM on two independent grounds, either sufficient: the general two-class problem is settled at the trivial bound, and the obstruction Tao names is our own configuration; the Birkhoff branch is subsumed and strictly worse from x = 13, and the row's payoff is a wall address rather than the published anchor the map priced. | none | [row12-recon.md](history/staging/row12-recon.md) |\n| `Q-row7-talagrand` | CLOSED | Does Talagrand's concentration on product spaces (IMPORT-MAP row 7) bound anything the corpus needs? | CLOSED WITH MECHANISM: the structural fit is better than the map wrote, since Theorem 8 needs no dependence structure and the shared-draw mechanism that closed rows 3, 4 and 10 does not close this one, but the per-coordinate effect c exceeds the largest tolerable value by a factor that DIVERGES (174x at @11 to 70,576x at @23), and the target hole is on the wrong side of the ensemble/anchor divide, so even a perfect bound would be inert. | none | [row7-recon.md](history/staging/row7-recon.md) |\n| `Q-scanstat-t37` | ANSWERED | What does the seventh exact level T37 do to the linear-in-ln D rule for H, and to the sqrt(m) refutation? | The sealed prediction H*5(T37) = 0.381254, band [0.369866, 0.392641], misses the measured 0.356548 +/- 0.0068 by 3.63 of its own s.e., so per the seal the linear form was a coincidence of a short ladder; the exponent series still rises but is concave in ln D, and G2(37#) is certified exhaustively over 217,929,355,875 slots. | none | [scanstat-t37.md](history/staging/scanstat-t37.md) |\n| `Q-scanstat-t37-prereg` | OPEN | What is the moving-sum exponent H at T_37, predicted before any producer for the pass existed? | Pre-registration only, sealed and committed alone: the prediction band, the criterion, the weak secondaries, the custody gates the producer must pass before reporting and the compute plan are fixed, and T_31 is explicitly excluded as another pre-registration's verdict. | none | [scanstat-t37-prereg.md](history/staging/scanstat-t37-prereg.md) |\n| `Q-scanstat2` | ANSWERED | Does the scan-statistic exponent law survive a sixth exact level, and is the tail factor the extremal index? | Both follow-ups come back negative: T_31 missed the pre-registered exponent rule from outside its own 95 per cent band, so the linear-in-ln D law describes five points and is not a law, and the tail factor is not the extremal index, its correction pointing the wrong way at seven of nine tested places; the sixth level confirms the kill on sqrt(m) harder, at H = 0.3460 +/- 0.0068 against 0.5. | none | [scanstat2.md](history/staging/scanstat2.md) |\n| `Q-scanstat2-prereg` | CLOSED | Does the linear-in-ln D exponent rule predict H(T31), and is the tail factor the extremal index? | Sealed and committed alone before any producer for the pass was written; scored in scanstat2.md, where both follow-ups come back negative: T31 missed the pre-registered band from outside, so the linear law describes five points and is not a law, and the tail factor is not the extremal index, with the correction pointing the wrong way at seven of nine. | none | [scanstat2-prereg.md](history/staging/scanstat2-prereg.md) |\n| `Q-search-conventions` | ANSWERED | In whose vocabulary do these objects live, and which prior-art searches have already been run in the owning convention? | The owning conventions are tabulated by object and the searches already run are listed so they are not repeated; the rule stands that a clean negative in our own wording proves nothing, and section 5 keeps what is still uncovered separate from what is covered. | none | [SEARCH-CONVENTIONS.md](SEARCH-CONVENTIONS.md) |\n| `Q-self-contradiction` | ANSWERED | Which documents in the corpus contradict themselves? | Twenty findings, two rated HIGH because they change what somebody does: theta-ladder.md's retracted crossing still asserted unmarked in three further sections, and G2-STATE.md's confirmed G2(41#) = 546 against three sections still carrying twelve terms; eight more change a number somebody quotes. | none | [audit-self-contradiction.md](history/staging/audit-self-contradiction.md) |\n| `Q-session-0828` | ANSWERED | What did the 2026-08-28 wave establish? | 21 agents, zero routes, four TODO items closed or answered, one HELD headline (Var/E = 0.45546, heuristic). | none | [session-0828-summary.md](history/staging/session-0828-summary.md) |\n| `Q-session-0829` | ANSWERED | What did the 2026-08-29 session (the Z/G object reads, the measurement wave, the red teams, the refuted-claims audit) establish, and what does it leave open? | Nothing moved an exponent or opened a route; the map moved: a Z2-STATE draft exists, both objects carry labelled legal-open-sets, the Gap Reformulation's discarded origin is proven not needed for the implication and measured worth one unit of exponent against a deficit of 2.2665, all 67 REFUTED rows re-derived with none reopening and 27 clauses corrected, and about forty live-layer and HELD-note defects were applied after verification. | none | [session-0829-summary.md](history/staging/session-0829-summary.md) |\n| `Q-session-0830-summary` | ANSWERED | What did the 2026-08-30 overnight campaign (rolling waves of up to ten agents on the queue rewritten after the 08-29 evening wave) change, and what is still open? | Neither exponent moved; twenty-one staging notes landed, all HELD; four routes closed (the skeleton door, the K*-product and fold-composition doubling bridges, the deep-ladder Buchstab transfer), TODO item 4 deleted and items D and 8 parked, the RML route carries an adverse derived floor whose growth law survived one sealed blind test and awaits a red team only Chris can call, Z2's fifth decade rejected the exact-Mertens main term for the omega-form crossing at 13 to 24 se, three record defects were verified on independent code and applied as riders, and a coherence pass corrected 27 lags in the live layer. | none | [session-0830-summary.md](history/staging/session-0830-summary.md) |\n| `Q-session-0904-summary` | ANSWERED | What did the 2026-09-04 wave (four Opus attacks on wall-facing questions, one orchestrator note, four Opus red teams, all verdicts applied) change, and what is still open? | Neither exponent moved; item 0's growth half survived a second adversarial pass so the route is a truth gap at rung derived-and-red-teamed-twice; Face 4's \"no lower bound on the kappa = 2 sifting limit is known\" was a convention failure and published floors at or below 2 are now cited, while the recon's headline that the cap is a method artefact was refuted by its red team; killer 2's coordinate is measured for the first time (argmax mirror-invariant, zero congruence pairs from x = 23, forced) after the multiplicity half turned out to be on the ladder already; no piece of R0 is both legal and parity-exempt with content; L7 is not a transfer; about thirty live-layer sentences corrected across seven files. | none | [session-0904-summary.md](history/staging/session-0904-summary.md) |\n| `Q-shadow-amplitude` | MIXED (shadow-amplitude-prereg.md: OPEN; shadow-amplitude.md: PARTIAL) | Where does the kill shadow's drift amplitude come from, and does the record's counting floor explain it? | The finite-y correction is an identity with no free parameter and both measurement routes agree 10 of 10, but the record's own counting floor is four to six times too small and its candidate explanation is the wrong half, so the label stays PARTIALLY EXPLAINED. | 5 (retired) | [shadow-amplitude-prereg.md](history/staging/shadow-amplitude-prereg.md), [shadow-amplitude.md](history/staging/shadow-amplitude.md) |\n| `Q-shadow-buchstab` | ANSWERED | Is the kill shadow the band-averaged pair-Buchstab integral? | SURVIVES WITH CORRECTIONS, and under the record's own pre-registration the verdict is SHAPE-ONLY rather than DERIVED, since y ~ 1000 misses D1 at 1.49x tolerance; the drift's amplitude is off by about 2.6x, the coefficient has the closed form 2 - 1/ln 2 = 0.5573049591 that the record missed, and anchored-windows section 5 reproduces 32 of 32 from an instrument sharing no code. | none | [shadow-buchstab.md](history/staging/shadow-buchstab.md) |\n| `Q-shadow-prereg` | ANSWERED | Is the kill shadow's 0.85 the band-average of the Unification-Law survival curve over the post-crystallization window? | Scored in shadow-buchstab.md: SHAPE-ONLY under the original y >= 997 rule; D3 passes in the reported ten clusters but D1 fails near y = 1000 at 1.49 times the 0.011 tolerance. Later finite-y normalisation and counting-floor corrections do not promote this verdict. The pair-Buchstab identification remains conjectural; the pre-registration body is preserved. | 5 (retired) | [shadow-prereg.md](history/staging/shadow-prereg.md) |\n| `Q-sharp-corner-transition` | ANSWERED | What is the energy of the complete sharp corner coefficient, and can the smoothed estimate be transferred through a negligible L2 transition? | DERIVED using named analytic inputs: for each fixed 00 both sharp squared norms and sharp-minus-smoothed squared norms are Theta_eta(x log^2 x) on the actual dyadic intervals. Thus an O_eta(x log x) sharp squared norm or negligible L2 transition fails in that range. Signed transition correlations and the global complement remain OPEN. No region, exact residual cut or twin margin changes. | C | [sharp-corner-transition.md](sharp-corner-transition.md) |\n| `Q-sharp-sieve-range` | CLOSED | Do the sharp sieve functions (Jurkat-Richert, DHR) make the two empty certificate-engine theorems non-empty at run levels? | NO at finite level: DH Thm 9.1 carries no written constant and the crude fundamental lemma is strictly better; as limit statements kappa=1 is first non-empty @37 (a 19.8-wide bracket; @53 is the first narrow one, 1.656) and kappa=2 @23. | 8 | [thm-sharp-sieve-range.md](history/staging/thm-sharp-sieve-range.md) |\n| `Q-shifted-prime-decomposition` | ANSWERED | Does decomposing Lambda(dk-2) give a provable saving on any part of the actual shifted-prime sum, and what exact arithmetic remains outside the imported hypotheses? | Both second Type I terms are O_H(x/log^H x) by classical Mobius BV. The residual is an explicit weighted sum over dk-ev=2, equivalently an average of two-linear-form Mobius correlations with growing coefficients and possibly one-point intervals. Its required one-sided improvement is OPEN. No twin lower bound or novelty is claimed. | C | [shifted-prime-decomposition.md](shifted-prime-decomposition.md) |\n| `Q-shifted-prime-mobius-sums` | ANSWERED | At reachable x, do the shifted-prime Mobius and Liouville sums, the cleanest instance of the parity input the twin reduction lacks, show any size or sign structure that a random-sign model does not, and how loose is the trivial bound on M(x) used for the centered tolerance? | MEASURED to x=2^38 (10.9e9 primes). All four sums sit at random-sign size: \\|U\\|/control rms at most 2.8 at any j, slopes 0.28 to 0.65 against the control's 0.43 to 0.47 with per-draw spread 0.19 to 0.30; the pre-registered structure falsifier did not fire. M(x)/x is at most 8.8e-5 in magnitude for j>=29 against the trivial bound 0.374, so the -4x/25 tolerance in moving-cutoff-parity is loose by four orders at these scales; a proof still needs a bound on M, which is the same parity object. The squarefree density of p+2 among primes is 0.747911 at 2^38 against 2*Artin = 0.7479116. pi(2^j) matches OEIS A007053 at every j. Nothing here bears on the asymptotic conjecture. | C | [shifted-prime-mobius-sums.md](shifted-prime-mobius-sums.md) |\n| `Q-sift-limit-discards` | PARTIAL | Where does the dimension-2 sieve discard our structure, and can any of it re-enter to beat beta2 = 4.2665? | Every asset checked against the five discard points has no known consumer and no entry point, and EH and GRH change nothing here; Lemma V's mean-square form is PROVED and hands back a sup bound, the strata are self-similar and it still does not close, L has no combination law of its own, and the 529 covering route is closed rather than unfinished. | none | [sift-limit-attack.md](sift-limit-attack.md) |\n| `Q-sigma31-calibration` | ANSWERED | Is the sigma behind the @31 X-channel detection calibrated at @31 itself? | Calibrated at its own level for the first time: seven known-truth control draws give sigma_true/sigma_reported = 1.127 at the point, 95 percent bracket [0.745, 2.293] (pooled with the earlier draws, [0.772, 1.664]), which restates the detection at z = -2.06 under the point calibration and closes both T2.b findings; the detection weakens rather than dies. | none | [attack-sigma31-01.md](history/staging/attack-sigma31-01.md) |\n| `Q-signed-divisor-grouping` | ANSWERED | Can grouping different fibers while retaining their Mobius signs control an explicit part of R, and what signed error remains after that grouping? | For every fixed H>0 the actual remainder restricted to de<=x^(7/10) is O_H(x/log^H x), although each sign's mass there is at least c*x*log(x) eventually. This follows from a written classical-input derivation, with finite algebra and archived-factor checks. The remaining de>x^(7/10) contribution reduces to an explicit weighted CRT endpoint discrepancy whose required one-sided improvement is OPEN. | C | [signed-divisor-grouping.md](signed-divisor-grouping.md) |\n| `Q-signed-moment` | PARTIAL | Is there an argument for the block that does not pay sqrt(M) at the first Cauchy inequality or that extracts cancellation from the R=0 class, and what are its budgets at the corner a=b=1 and at the target box (delta,nu)=(8/25,9/20)? | The true diagonal has an upper bound x^(1+o(1)) on the transition band, not a uniform nonzero lower bound. Lemma A controls only u1=u2,h1=h2; unequal proportional R=0 pairs remain outside it. Lemma B and the now-priced joint Cauchy arrangement add no region. The former general Holder floor is valid only under additional coefficient assumptions, not for arbitrary sparse coefficients. Conditional off-diagonal budgets and the random-matrix ceiling remain conditional and heuristic. No necessity of using both signs, full-corner estimate or sufficient twin margin is established. | C | [signed-moment.md](signed-moment.md) |\n| `Q-single-alignment` | CLOSED | Does the head's single-alignment fold recursion for M(x, x^3) give a proven-shaped handle on the per-fold multiplier that the max-over-alignments recursion lacked? | The answer splits: the multiplier IS different in shape, measured - exactly 1 at 225 of 240 folds from x = 53, total spend 1.09 nats against a tile-shaped ladder's 7.12, measured M(1613) = 2220 against a tile-shaped 3.7e5 - but the route to a growth law through the recursion closes anyway, because the growth is initiated by the window's expansion into fresh ground, a boundary term that is the localized gap problem re-posed and that no fold reaches. | 0d (retired) | [localized-single-alignment.md](localized-single-alignment.md) |\n| `Q-singleton-fiber-audit` | ANSWERED | Can the one-point fibers of the coupled remainder be discarded or controlled in absolute value at scale x, and what do the actual weights and archived intervals show? | The grouped coefficient is zero whenever either integer is prime. Each sign's ungrouped singleton mass is at least c*x*log(x) eventually for an absolute c>0, by a written derivation from named classical inputs. Finite full-interval and archived-prefix checks validate the partition. The signed singleton and longer-fiber contributions remain OPEN; no twin lower bound or novelty is claimed. | C | [singleton-fiber-audit.md](singleton-fiber-audit.md) |\n| `Q-skeleton-bound` | PARTIAL | Does the aggregate 30-skeleton bound hold, and can it be certified at every level? | Certified as an exact BigInt inequality at six levels @11 through @29, margins 0.287 to 0.406, with the Skeleton Collapse Theorem proven for all x and all q and an exact exception criterion; the all-x form is open and its blocking term is prime-equidistribution of window residues rather than branch algebra. | none | [natal-cap-30-skeleton-bound.md](natal-cap-30-skeleton-bound.md) |\n| `Q-skeleton-decide-0830` | ANSWERED | Should TODO item 4, \"Skeleton Equidistribution: restate or demote\", be restated around the modulus-W mass or demoted off the board, and what exactly is banked either way? | DEMOTE. Over every scour prime the branches on which the door is a fixed-modulus question carry 9.2, -0.9, 0.2 and 0.5 percent of the skeleton at @13, @17, @19, @23 (the -10.8 and 5.5 percent on record were 9 and 2 percent subsamples), so the door as named removes at most 0.0102 from a G30_agg whose open part is 0.094 to 0.126; on the open side the phase never wraps and no equidistribution statement remains, only the inequality; nothing on the live board consumes G30_agg < 1/2; no first move exists that is not already ANSWERED. Banked regardless: the Skeleton Collapse Theorem (PROVEN, all x, all q) and G30_agg < 1/2 CERTIFIED at six levels @11..@29. | 4 (retired) | [decide-0830-skeleton-door.md](history/staging/decide-0830-skeleton-door.md) |\n| `Q-skeleton-door` | ANSWERED | Is the Skeleton Equidistribution Conjecture the blocker on the anchored calm? | It is not the blocker as far as the door can be seen: Theorem A (Trapezoid Cancellation) is proven for all x, q and branches, @29 is certified as an exact integer inequality at G30_agg = 0.1176 with margin 0.3824 over 7,863 scour primes, and the decay-law shortcut is REFUTED, the six levels being flat within their own spread with every fit made strictly worse by adding @29. | 4 (retired) | [natal-cap-36-skeleton-door.md](natal-cap-36-skeleton-door.md) |\n| `Q-small-divisor-kernel` | ANSWERED | Does reciprocity separate the j<=x^(1/20) kernel of grouped-divisor-moment (21) at delta=8/25, nu=9/20, and do the Duke-Friedlander-Iwaniec / Bettin-Chandee and Kuznetsov interfaces then supply the required saving? | The separation is exact and cheap: reciprocity splits index by index, the correction phase has sup norm and total variation O(A/(MN))=O(1/x) at this box, and Mellin separation of the endpoint weight costs x^epsilon with L1 mass min(1,Ax/(MN)), so the whole block reduces to a pure trilinear Kloosterman fraction with unimodular twists. The interfaces then fail by explicit amounts: Bettin-Chandee Theorem 1 gives block exponent 129/125, above the required 1 by 4/125 and above the classical 103/100; DFI (1.1) gives 1267/1200; every box Bettin-Chandee controls is already controlled, verified on 19005 rational boxes; the audit's spectral diagnostic factor is trivial for all common divisors up to x^(19/50). Closing this box inside the same interface needs the (M+N) exponent below 27/140 or the (AMN) exponent below 9/28. Target (21) and the global twin margin remain OPEN; nothing is added to the controlled region. | C | [small-divisor-kernel.md](small-divisor-kernel.md) |\n| `Q-smooth-sieve-literature` | PARTIAL | Which nearby published results should be imported instead of rediscovered, and do their methods give a more focused next question for the complete global remainder? | Existing sources supply the finite-difference mechanism, one-point concentration and quadratic profile optimization; shifted divisor-sum estimates admit fixed shift two under their own support and smoothness conditions. A derived application of corrected Henriot bounds gives small-prime mass O_sigma(delta^sigma x)+O_epsilon(x^(39/40+epsilon)), uniformly in delta for fixed 0floor(x^(77/100)) as a fourth simultaneous condition of W_dagger through the Perron machinery, changing only the summation domain of E_dagger. The target box (407/400), the corner, the uniform product threshold 19/25, the sign of E_dagger and the global twin margin are unchanged; the margin remains OPEN. | C | [structured-dispersion-estimate.md](structured-dispersion-estimate.md) |\n| `Q-supported-coefficient-dickman` | ANSWERED | Can the whole supported coefficient be priced using existing literature before enumerating its factor cells, and does the resulting separate-sign sieve estimate supply a positive twin margin? | DERIVED from classical inputs: at delta=1/100 the harmonic limit is H_delta=D(25/12)-1+E_delta with 0<=E_delta<=24/22!, hence -13/2, where I is the explicit ordered two-factor integral below. A three-prime left input and rough composite right input suffice; prime partners are subtracted using an upper sieve. The witness survives every fixed delta<1/50 and every averaging profile with the same plateaus. Consequently the negative-only consumer is false, and liminf Phat/(C2*x)>1/2. The signed twin margin remains OPEN; literature novelty is unestablished. | C | [switching-negative-mass.md](switching-negative-mass.md) |\n| `Q-tail-deficit-model` | ANSWERED | Does the exact compound-geometric thinning of the previous level's own gap histogram, with the qualifying-gap constraint g = 0, +/-2 (mod q) carried through the ladder, predict the tile's tail-deficit factor G2/(mbar ln D) at zero parameters, and does that make object-models-read-0829's D5 residue-level classification constructive rather than inferred? | NO. The zero-parameter CRT-constrained thinning predicts a tail HEAVIER than the tile's at every level, delta_M2 = 0.6461, 0.6746, 0.7021 against a measured 0.4577, 0.4463, 0.4791, so the pre-registered \"classification was wrong\" arm fires and D5 returns to OPEN with C3's constructive route REFUTED; the constraint is informative but small, carrying 4.0 to 10.0 percent of ln(null/truth) at the far tail, or 7.7 to 14.4 percent read against a rate-matched control, against the grain's ORDER at 39.4 to 54.6 percent at the deepest abscissa each level resolves; at shallow abscissae the ordering reverses (B 8.7 against C 4.8 at x = 23, u = 4), which the exact operator on a shuffled true grain confirms without any model (maximum 276, 354, 426 against the true 204, 258, 348), and the order effect is now localised (sections 7 and 9): a first-order Markov word from the true adjacent-pair statistics carries 32.3, 34.2 and 60.8 percent of ln(shuffled/truth) at the deepest abscissa of the three folds, and a SECOND-order word from the true adjacent-triple statistics carries 97.3, 97.9 and 102.6 percent there, r = 0.972, 1.061 and 1.000 on draw-mean maxima over 40, 20 and 10 draws, so the folded tile's tail is MEASURED to be a function of the previous grain's three-consecutive-gap statistics and of essentially nothing longer-range; that surrogate is seeded on the true previous grain at one fold, is not iterated, predicts no delta and does not revive C3, and the 348.0 sd 0.0 at 29->31 rests on ten draws. | 0 | [measure-tail-deficit-0829.md](history/staging/measure-tail-deficit-0829.md) |\n| `Q-tail-derivation-0830` | PARTIAL | Does the tail's measured law c = 0.7522 ln^2(p'^2) and its surplus derive, and which part of the coefficient is the tile's own (no prime input) against which part needs Hardy-Littlewood? | PARTIAL, kill clause fired on the coefficient: the ensemble object (the same functional averaged over all p# translates) derives exactly, E_all = R + 5/2 and E_odd = R + 3 with R = Sum g^2/2W, computed exact at x = 7..29 with every conditional the rough-origin candidate needs, and its Mertens constant e^{2gamma}/(8 C2) = 0.6007 is 1/(2 C2) times rho(2) = e^{2gamma}/4 by construction; every route from there to the anchored 0.7574 passes through rho(2)'s limit (the sharp form of Assumption A, algebraically equivalent to HL) and through the zone's gap shape CV^2 -> 1 (an HL statement), so the anchored coefficient is HL in disguise and nothing beyond HL is derived; the tile's own (1 + CV^2)/2 is 0.6306 -> 0.7532 over x = 7..29 and not settled; the record's 0.7522 = HL x 0.9644 x 1.0298 is two non-HL-sized factors cancelling at this height; the rough-origin candidate has an exact ensemble counterpart E_rc/E_cls = 1.0050 -> 1.0523, right sign, wrong size and trend against the anchored 1.0157, not identified; one number derives that the record measured: the class-null offset E_cls - R = 5.600 -> 6.034 exact against the record's 6.05, replacing both the prereg's 0.5 and the note's 7.5; and the corpus's \"R + 1/2\" is the other convention, under the tail's own the constants are 5/2 and 3. | Z4 | [attack-0830-tail-derivation.md](history/staging/attack-0830-tail-derivation.md) |\n| `Q-tail-maximal` | CLOSED | Is a tail-only maximal inequality a smaller ask than the full one, as attack-AB-bounded section 4.4 supposed? | No: the split is exact but is a reparametrisation rather than a localisation, the tail is maximal and saturates the Gaussian law, a fixed cut buys nothing asymptotically, and the tail is not a smaller ask but the whole ask. | none | [attack-tail-maximal.md](history/staging/attack-tail-maximal.md) |\n| `Q-tailcount-transport` | ANSWERED | Does the Tail-Count Transport survive an independent adversarial re-derivation? | The inequality, the certificates and the NO-CHAIN verdict all STAND on an implementation written from the operator definition, with a seventh fold added exact (31: 348 = 348); three corrections follow, the stated equivalence is false at fold 29, zero loss is exactness-by-relaxation rather than a tail fact, and the measured spends are 1, 1, 2, (1 or 2), 3, 2. | none | [verify-tailcount-transport.md](history/staging/verify-tailcount-transport.md) |\n| `Q-tau-repricing` | CLOSED | Does the tau(m) re-pricing of the x-free bound stand? | The separability claim is true and re-derived four ways, but the re-pricing substitutes one quantity for a different one: summed over the modulus set the tau(m) bound IS the divisor-pair count up to the vector sieve's fixed factor 3, and used as a bound it is 4 to 5x weaker than the exact quantity in the Fourier basis and 23 to 44x weaker in the divisor-pair basis. | none | [attack-tau-repricing.md](history/staging/attack-tau-repricing.md) |\n| `Q-the-dials` | ANSWERED | Which variables can this programme actually push, and which is Chris missing? | The square frontier constrains certification from roughness alone, not every arithmetic method. The current campaign uses long intervals, a classical Chen benchmark and an OPEN signed estimate; uniform maximum-gap control is a sufficient secondary target. Finite slopes do not determine an asymptotic exponent. | none | [THE-DIALS.md](THE-DIALS.md) |\n| `Q-the-lens` | ANSWERED | Why is the tile the right object, and what does the lens buy? | Completeness is PROVEN and elementary (every twin prime above p is a twin slot of T_p) and the recursion is exact; nothing here is claimed as new mathematics, and the triage rule of section 5 states what the lens does not buy. | none | [THE-LENS.md](THE-LENS.md) |\n| `Q-theta-ladder` | CLOSED | Does the all-positions exponent theta turn over below 2, converge to exactly 2, or settle above 2? | The answer is above 2 with an amendment, and the route is retired regardless: theta does not turn over (it dips and comes back, need/z^2 = 0.3550 to 0.6119 at z = 13..31 and rising from 31), convergence to exactly 2 is not supported (e = 1 rejected at 95% in all four fits), point estimates run 2.37 to 2.99 without settling, and OUTCOMES.md records the certificate route as RETIRED because the sharp maximal law it needs is itself TPC-implying. | 0 | [theta-ladder.md](theta-ladder.md) |\n| `Q-theta-last-gap` | CLOSED | Can the last 0.209 of theta_total be recovered by sharper absolute-value accounting? | No: theta_total stays 1 and the obstruction now carries an exponent, the absolute-value ceiling is c*H^theta_total with c > 0 rather than H*polylog, so the shortfall is a power H^0.212 and the entire budget any constant could ever buy is O(lnln H / ln H) against a break-even of 1.208983. | none | [attack-theta-last-gap.md](history/staging/attack-theta-last-gap.md) |\n| `Q-theta-selfconsistent` | ANSWERED | What does the exact-supremum theta ladder give when re-run self-consistently in H? | All five published self-consistent values reproduce exactly from two independent implementations; z = 31 is now measured at nP = 588, nP/z^2 = 0.6119, so the published 1.1776 is overstated by 1.93x, continuing the pattern, and at z = 47 a prefix walk finds the certificate failing at H = 1.0412 z^2, so the self-consistent unconditional requirement leaves the zone budget by z = 47. | none | [theta-selfconsistent.md](history/staging/theta-selfconsistent.md) |\n| `Q-transition-energy-review` | ANSWERED | Does the sharp-energy chain in sharp-corner-transition.md survive an adversarial check, and do the upper norm, the small-fixed-eta lower norm and the failed negligible transition each stand at their stated scope? | All ten assigned rows survive at their stated scope; no counterexample or decisive contradiction was found. The upper norm, the small-fixed-eta lower norm and the refuted negligible L2 transition each stand separately. Two repairs are citation-level only: the integer Barban-Vehov input is Graham 1978 quoted by Chen An, and the note should record why the k=Q constant is absolute. The de la Breteche-Dress-Tenenbaum PDF was refetched and its SHA-256 matches the pinned hash; its (1.5) and Theorem 1.1 supply exactly the imported (3) and (4). eta_0 remains positive and unextracted, and the signed sums, E_out and the twin margin remain OPEN. | C | [transition-energy-review.md](transition-energy-review.md) |\n| `Q-transition-energy-review-spec` | ANSWERED | What independent checks are required before the sharp-energy and non-negligible-transition conclusions can guide the next signed research attempt? | Bounded review assignment with explicit source, algebra, interval, error and quantifier checks; per-claim verdicts and dependency consequences are required. The assignment is specified, not executed. | C | [transition-energy-review-spec.md](transition-energy-review-spec.md) |\n| `Q-transition-joint-budget` | PARTIAL | Can the mixed and smoothed terms be handled jointly with the transition pair, and what complete inequality would make such an estimate useful for the twin consumer? | The exact split, cutoff-average identities and separate upper budgets survive. Averaging cutoff norms after triangle and Cauchy is saturated for sufficiently small fixed eta, but signed cutoff arguments and joint cancellation are not closed. R_00 need not be estimated separately under every grouping. The short-window constraint is corrected with a bounded majorant and a safely reduced lower constant. No consumer input or signed improvement is supplied; E_out and the sufficient twin margin remain OPEN. | C | [transition-joint-budget.md](transition-joint-budget.md) |\n| `Q-transition-joint-budget-spec` | ANSWERED | Can the mixed and smoothed terms be handled jointly with the transition pair, and what complete inequality would make such an estimate useful for the twin consumer? | Bounded assignment to audit the exact four-term decomposition, price a concrete joint treatment, and return a sufficient one-sided or common-scale inequality with all open terms explicit. The decomposition and existing norm budgets are inputs, not new findings. | C | [transition-joint-budget-spec.md](transition-joint-budget-spec.md) |\n| `Q-transition-research-execution` | ANSWERED | How should the core research handler execute and integrate the three bounded transition assignments while retaining the central signed estimate? | Prepared execution contract with a fixed mathematical baseline, three independently assignable specs, handler-owned signed research, dependency checks, shared-file ownership, reporting requirements and stopping rules. Execution is complete; the corrected results are in transition-round-review.md and transition-round-audit.md. The specification itself supplies no arithmetic estimate. | C | [TRANSITION-RESEARCH-EXECUTION.md](TRANSITION-RESEARCH-EXECUTION.md) |\n| `Q-transition-round-audit` | ANSWERED | Which conclusions of the completed transition round survive direct review, which require correction, and what research decisions remain justified? | The sharp upper norms and sufficiently-small-fixed-eta lower norms survive. The integration contained substantive errors: an omitted lower cutoff, an incorrect Mellin formula for the transition, a conflation of singleton divisor fibers with long cofactor fibers, an unpriced source representation and overbroad closures of joint or cutoff arguments. Owning notes and current records are corrected. No signed improvement or sufficient twin margin is established; the corrected cofactor and joint-correlation interfaces remain open. | C | [transition-round-audit.md](transition-round-audit.md) |\n| `Q-transition-round-review` | ANSWERED | What survives the executed transition round after direct review, and what combined estimate is still missing? | The energy chain survives at its original scope; the integration's missing cutoff, Mellin representation, fiber identification, normalization and general closure claims are corrected. The exact cofactor and joint-cutoff formulations remain useful. No signed estimate, controlled region, exact cut or sufficient twin margin changed. Only the stated norm-only procedure and the arbitrary-factor fixed-shift relaxation are refuted at their specified scopes. | C | [transition-round-review.md](transition-round-review.md) |\n| `Q-transition-signed-estimate` | PARTIAL | Can the signed transition kernel (E1) be estimated beyond its absolute norm budget, and if not, what exact input does the consumer need? | No signed improvement. Derived: the proper-prime-power part of R_11 is O(x^(39/40+epsilon)); the prime part is an exact beta-weighted sum over pairs of linear forms of two-point Mobius correlations with coefficients up to x^(6/25+2eta) and x^(1/20+2eta); the Cauchy budget is at most 2B*eta*x*log^2(x)(1+o(1)) and the fiber triangle budget 4*eta^4*x*log^4(x)(1+o(1)), so every existing inequality loses one of the two signs. The cutoff-difference identity converts T at pn into a sharp-window coefficient times log(p)/L plus two boundary windows; the entropy-decrement route then needs bounded coefficients, a short-interval one-point input for T and a log-weighted consumer, none available. The required one-sided input is stated exactly in section 5. The twin margin remains OPEN. | C | [transition-signed-estimate.md](transition-signed-estimate.md) |\n| `Q-transition-source-match` | ANSWERED | Do the inspected divisor-correlation, shifted-convolution and averaged-Chowla statements directly supply a usable estimate for the exact transition kernel? | No direct supplier was established among the inspected statements. The cofactor exchange and Theta identity are exact, but their scalar coefficient mass is not a full bounded-class representation for T. The transition Mellin kernel requires a sharp subtraction with a 1/s tail. Fixed-divisor singleton fibers differ from potentially long cofactor fibers. The fixed-shift arbitrary-factor relaxation is refuted; broader signed and cutoff methods remain open. | C | [transition-source-match.md](transition-source-match.md) |\n| `Q-transition-source-match-spec` | ANSWERED | What bounded source investigation could supply a usable signed estimate for the transition kernel, and how must its hypotheses and losses be priced? | Specification for up to three detailed source-interface assessments, including coefficient representation, scale quantifiers, normalization, separation errors and exact downstream payoff. No new theorem match or literature-absence result is claimed. | C | [transition-source-match-spec.md](transition-source-match-spec.md) |\n| `Q-twin-reduction` | ANSWERED | What has the arithmetic campaign of 2026-09-05 and 2026-09-06 actually established about the twin sum, at what calibration, what exactly remains, and what would suffice? | Consolidates the checked regional reduction S=C2*x+E_dagger+O_H(x/log^H x), its sufficient unbounded-scale margins, coefficient definitions and source dependencies. Later complete smooth and centered formulations are owned by RESEARCH-HANDOFF.md and their linked derivations. No new arithmetic estimate is supplied. Imported theorem proofs retain their stated review limits; every sufficient twin margin remains OPEN. | C | [TWIN-REDUCTION.md](TWIN-REDUCTION.md) |\n| `Q-two-class-lower-bound` | PARTIAL | What does the two-class LOWER bound on G2 say, and does it conflict with attack E's measured law? | The finite range cannot decide and it is not close: with both constants set to 1 the transferred bound and attack E's measured law do not cross until x = 10^7327 while the exact ladder stops at 79, so neither result refutes the other and no computation ever will. | none | [attack-lower-bound.md](history/staging/attack-lower-bound.md) |\n| `Q-two-class-lower-bounds` | PARTIAL | How large can G2(x#) be FORCED to be? | The best PROVEN lower bound costs nothing, the Kalmynin-Konyagin substitution is carried out here (DERIVED, independently checked, not refereed), and the certificate ladder is MEASURED with an oracle test whose load-bearing caveat is stated on its own; against the target x'^2 and the proven x^(4.2665+eps) upper bound the question stays open. | none | [two-class-lower-bounds.md](two-class-lower-bounds.md) |\n| `Q-two-moire` | ANSWERED | Do the Grain and the Scour, generated by disjoint prime alphabets, force twins to be infinite? | The provable content is exact CRT factorization: over the JOINT period the miss count is guaranteed positive, so aggregate alignment is arithmetically impossible, but that is an aggregate statement over a period vastly wider than the zone and does not reach TPC; the nearest unclaimed theorem is a two-class Jacobsthal upper bound at ANY exponent, and exponent < 2 would prove TPC outright. | none | [two-moire-argument.md](two-moire-argument.md) |\n| `Q-u-frame` | PARTIAL | Does restating the Zone Postulate as one inequality in u, stepping on odd numbers rather than on primes, buy anything against the wall? | It buys form and closes off a class of dead ends, but no difficulty and no slack: the requirement ln c(p) <= 2 ln p/p is exactly what the per-prime analysis already gave and the measured ladders run at about the full rate, so this is a change of coordinates rather than of terrain. | none | [U-FRAME.md](U-FRAME.md) |\n| `Q-u2-engine-depth` | ANSWERED | Is the certificate engine's operative depth heading to u = 2, and does that make TODO item 8's payout form TPC-strength? | It is not: the measured operative depth runs u = 4.191 at @17 down to 3.557 at @97 and is still falling, so the brief's conditional does not fire and item 8 is shown neither TPC-strength nor safe; a WALL-ADDRESS in the weakest sense, every figure SCRATCHPAD-GRADE. | 8 | [u2-engine-depth.md](history/staging/u2-engine-depth.md) |\n| `Q-unreached-sources` | ANSWERED | What is actually in the three sources SEARCH-CONVENTIONS section 5 lists as unreached? | The 2009 SeqFan thread on A144311 does not exist, now a read negative over 19,964 archived messages with zero hits; Paseman's n is the number of distinct prime factors and at that reading his shape is asymptotically weaker than our x^{4.2665}; MathSciNet stays paywalled and unswept. | none | [audit-unreached-sources.md](history/staging/audit-unreached-sources.md) |\n| `Q-usup-convention-0830` | ANSWERED | Are lemmaV-sup-extension.md's u_sup and u_sat and attack-0829n-rml-proof.md sec.4.1's CAP(z) statements about the same object (level D = z^s at s = 3.0, same moduli, weights and normalisation), so that the sec.4.1 correction in passing applies, or does a level-convention mismatch void it? | SAME object, VERIFIED at the code: both notes build the lattice through buildTerms(z, z^3), the modulus sets and Vabs(e) agree to 6.9e-18 at z = 13..23 and the count convention matches at all ten levels, the cited Ssat and u_sat reproduce on the RML lattice to every printed digit at z = 13..19 and the full-level alternative does not (19.544 against 19.602 at z = 13, 45.827 against 50.314 at z = 17); so the PROVEN cap Ssat <= CAP <= z^{2s+o(1)} applies and refutes lemmaV-sup-extension.md's asymptotic prose at lines 485-486 and 508-516 (2^pi(z) moduli, a C^pi(z) theorem as the reachable end), while sec.4.1's attribution sentence overreaches by calling that note's \"reading\" a shape it lists as one of two indistinguishable fits and flags as its likeliest error; neither ledger verdict line changes. | 0 | [verify-0830-usup-convention.md](history/staging/verify-0830-usup-convention.md) |\n| `Q-var41` | OPEN | What does the stable law predict for Var(41), and what can the tenth Var/E point pin? | Pre-registration only, sealed and committed alone before any Var(41) engine exists: it freezes the prediction, a band taken from the law's own residuals at z <= 37, the derived z(41) prediction, and the honest statement that one more point cannot separate a limit from a drift. | 2, 9 | [var41-prereg.md](history/staging/var41-prereg.md) |\n| `Q-varE-identification-0830` | PARTIAL | Can either open step behind lim Var/E = 0.45546 (the identification delta*(X - X_dec) -> 0, or the theta = 2 mean-coefficient replacement) be proven, and if not, which single inequality does not close? | Neither closes, and the two named steps are one statement (Conjecture 1 of variance-note sec.10; the model half is PROVEN in varE-limit-theorem.md). The remainder is re-split exactly at five levels: the corpus's X2 column doubles the positive half of one shift pattern (X2c = 2.66 against the group sum 0.09 at x = 19, PROVEN by rebuild), so its Xmix is overstated 2.3x; the true mixed remainder is MEASURED at -0.49 ln y and 82 % of it sits BELOW 2L, in the two-branch type, in the unbalanced range min(d,e) <= L^(2/5) (99.7 % of that cell at x = 19), outside the range of every bilinear Kloosterman-fraction bound; the divisor-distribution estimate the record names (n > 2L only) is therefore not the whole open step. Henriot's Corollary 2 (read at the page, erratum read) applies as stated and cuts the moduli to n <= L ln^(2+o(1)) y, PROVEN given that theorem; the one inequality left is the uniform o(1) equidistribution of y-friable squarefree integers in progressions to moduli up to y^(4/5), NOT SEARCHED at the page. The limit 0.45546 stays HEURISTIC. Not TPC-strength. | 9 | [attack-0830-varE-identification.md](history/staging/attack-0830-varE-identification.md) |\n| `Q-varE-limit` | PARTIAL | Does lim Var/E on the diagonal window exist, and what is it? | The theta=2 mean-coefficient step is neither proven nor refuted: the replacement error is exactly a sum over shifts of W(h) - V(h), it splits into a c=0 group (needs no decoupling) and CRT-mixed lags (open); measured ratios true/model 1.0098, 1.0039, 1.0014, 1.0013 at x = 13..23 with delta(X - X_dec) ln y falling rather than settling, so the error is O(1/ln y) or better MEASURED on four levels that exclude growth and nothing finer, and the limit 0.45546 stays HEURISTIC with varE-spectral's second step, its own limit theorem, still open. | 9 | [lit-dickman-variance.md](history/staging/lit-dickman-variance.md), [lit-smooth-divisors.md](history/staging/lit-smooth-divisors.md), [varE-asymptotic.md](history/staging/varE-asymptotic.md), [varE-exact-ladder-01.md](history/staging/varE-exact-ladder-01.md), [varE-limit-theorem.md](history/staging/varE-limit-theorem.md), [varE-spectral.md](history/staging/varE-spectral.md), [varE-theta2-proof.md](history/staging/varE-theta2-proof.md), [varE-theta2-step.md](history/staging/varE-theta2-step.md) |\n| `Q-vc-prior-art` | ANSWERED | Does arXiv:2208.06442 contain or overlap the VC-dimension = 4 finding of import-vc-nets? | DISJOINT at theorem level; not found there nor in the reachable one-hop neighbourhood; Helmbold-Sloan-Warmuth 1992 Thm 3.1 owed (abstract only). | none | [lit-vc-multiples.md](history/staging/lit-vc-multiples.md) |\n| `Q-verify-audit-numbers` | ANSWERED | Is audit-numbers.js circular? | Not circular in the way that would matter most, since it reads no other repo script, but 15 of its 117 checks certify nothing and 17 more are artifact echoes among transcribed constants; the one bug class the charter was written against had nothing in sections A to W to fire against and still has nothing, and one from-scratch re-implementation found a check depending on an unnamed hard-coded phase choice. | none | [verify-the-verifier-numbers.md](history/staging/verify-the-verifier-numbers.md) |\n| `Q-verify-cofactor-convolution` | ANSWERED | Does the cofactor-convolution identity at the anchor hold, and does it test the joint law? | RESTATED with corrections: the arithmetic is right, an independent re-derivation reproducing every figure to the digit at all five levels and confirming the asserted closed-form step, but the framing does not survive, since X is pinned by the identity X = M - Nbar + n_0 so the whole comparison collapses to one cell and the joint is measurably not a product elsewhere; the model error is 0.52 per cent, not 0.050. | X | [verify-cofactor-convolution.md](history/staging/verify-cofactor-convolution.md) |\n| `Q-verify-monotone-depth` | ANSWERED | Does the Monotone Depth obstruction, which retires dial 4, survive an adversarial re-derivation? | NARROWED: every measurement in the record reproduces, several to the last digit, but the mechanism as named is false as a general law and the corpus already holds a counterexample to it; what survives is an accounting of three channels priced in tenths of a nat against a need of 6.700, 8.191 and 9.637 nats, so dial 4 becomes priced slack with one channel open and unpriced. | 0e (retired) | [verify-monotone-depth.md](history/staging/verify-monotone-depth.md) |\n| `Q-verify-record-defects-0830` | ANSWERED | Do the three defect claims of 2026-08-30 (attack-0830-varE-identification.md sec.8 on the doubled X2 column; attack-0830-tail-derivation.md sec.7 on the corpus's \"R + 1/2\"; attack-0830-head-remainder.md sec.1 on the decade-pooled Delta = 0.6214) reproduce on independent code from the definitions, and which of the corrections they list apply? | Claim 1 CONFIRMED: the corpus delta*X2 column is the positive half of the p \\| h-2 pattern doubled, that pattern's mirror is the p \\| h+2 pattern (W-(-h) = W+(h) exactly, W-(h) != W-(-h) at 9 to 900,679 shifts), the group sum 2X2 reads -0.5603 .. 0.1846 at x = 7..19 against 5.7147 .. 5.3266, and Xmix is overstated 2.305x at x = 19 (6.9x at x = 7); X, X1 untouched. Claim 2 AMENDED: R + 5/2 and R + 3 are exact for the tail convention (all and odd origins, asserted at x = 7..23) and t/(R + 3) = 1.0228, but head-residual-factor.md:70 is the HEAD's forward convention where R + 1/2 and R + 1 are exact, so that correction does not apply; and p'^2 is 1 mod 6, where the tail constant is 5 (t/(R + 5) = 1.0181), with the class null (1.0157) the matched figure, so 1.0228 is one unmatched convention replacing another. Claim 3 CONFIRMED as an artefact, AMENDED on mechanism and on the half-decade reading: 0.6214 pooled against 0.6592, 0.6705, 0.6736 at 2, 4, 8 sub-windows; the between-slope (Simpson) piece the note derives is half of the pooling term (0.0187 of 0.0378 at halves, 0.0261 of 0.0522 at eighths), the other half is the within slope's own fall with height weighted by Var(g); and re-reading the halves from their eighths puts HL BELOW the measurement at 5 of 6 half-decades by 2.5 to 3.4 percent at the top decade, so \"within 1.6 s.e., sign alternating\" is itself a pooled statement. | Z4, 9 | [verify-0830-record-defects.md](history/staging/verify-0830-record-defects.md) |\n| `Q-verify-the-verifier` | ANSWERED | Does the QC gate's alarm ring when a defect of each real class is present, rather than merely reading green? | Of the 31 blind mutations the existing selftest covers zero: it is a reachability test proving each alarm can ring, never a sensitivity test asking how far a defect must move before the alarm stops, and its fourteen controls are all obvious silences; the ranked blind spots are listed with a proposed fix each, led by out-sha256 being written and never read. | none | [verify-the-verifier-checks.md](history/staging/verify-the-verifier-checks.md) |\n| `Q-w1-singular-series` | ANSWERED | What is the endpoint comb W1 in the literature's terms? | W1 is exactly the Hardy-Littlewood singular-series ratio for (0, 2, v, v+2) over the twin series squared, truncated to the folded primes and normalised by a v-independent constant, PROVEN algebraically and VERIFIED as exact BigInt-rational equality, so the fold-factor comb is the twin-twin correlation comb evaluated at v = theta_p. | none | [w1-singular-series.md](history/staging/w1-singular-series.md) |\n| `Q-width-repairs` | ANSWERED | Can the five loaded containers found by the width sweeps be guarded without invalidating any embedded figure? | Forward protection only: every repair is a widening or a bound check, storage-only by construction, no corpus figure was retroactively corrupt and none moved, with maxgap-law.js at 0 of 1389 figures unreproduced and attack-foldL-01-census.js byte-for-byte on the same out-sha256. | none | [width-repairs.md](history/staging/width-repairs.md) |\n| `Q-width-sweep` | ANSWERED | Which big-level enumeration and census scripts carry a fixed-width container that is safe at every level run and aliases silently above a threshold? | Three instances: two CORRUPT (the one-word residue mask that corrupted the census from x = 37 up, and a Uint16 index into the scour-prime array in the @37 census, both since fixed) and one that threw rather than aliased; every internal identity in the corrupted producer is a COUNT identity and stayed PASS throughout, because the alias corrupts which prime, not how many. | none | [width-sweep-census.md](history/staging/width-sweep-census.md) |\n| `Q-width-sweep-attacks` | ANSWERED | Does the attack and analysis script family carry the fixed-width-container defect class that corrupted the @37 census? | No live corruption across 123 scripts, every stored value derived at the level actually invoked and fitting its container; one hazard is serious, attack-x-offset-02-profile.js storing an index into the scour-prime array in a Uint16Array, the identical mechanism in the identical role as the census defect fixed at 605ce83, and it fires at @37. | none | [width-sweep-attacks.md](history/staging/width-sweep-attacks.md) |\n| `Q-wrap-identity` | ANSWERED | Can T4 be computed without enumerating quadruples? | Yes: the 4-point wrap identity exists and is verified through @13 against the certified value, turning @17's dead T4 (4.9e16 quadruples) into an 11-minute computation, but at a precision that certifies T4 itself and not yet the quartic bound. | none | [natal-cap-32-wrap-identity.md](natal-cap-32-wrap-identity.md) |\n| `Q-wrongdirection-audit` | ANSWERED | Are any of the ten live targets secretly TPC-strength, before the sessions are spent on them? | The audit closes no target: two of the ten come out TPC-strength and three more carry a TPC-strength face under a quantifier they are likely to drift into, which is a labelling result and not a reason to drop them; item 1e is settled only because the answer was already on disk, and one route-blocking number in it is hand arithmetic and unstamped. | 1e (retired) | [attack-wrongdirection-audit.md](history/staging/attack-wrongdirection-audit.md) |\n| `Q-xchan-at29-prereg` | OPEN | Does the joint-deficit closed form survive a blind test at @29? | Pre-registration only, committed alone before any producer existed: the statistic, the predictions adopted verbatim from the record, two acceptance bands, the validation gate the instrument must clear before any @29 number is reported, and what each verdict does to TODO item X. | X | [xchan-at29-prereg.md](history/staging/xchan-at29-prereg.md) |\n| `Q-xchan-at37-offset` | ANSWERED | Does any registered offset-correction candidate for the ~3.8 law survive at @37? | Sealed and committed alone before any @37 census of any kind existed, fixing the candidates, the sigma model and its projection band, the scoring rule and what each verdict does to item X's offset clause; scored in xchan-at37-score.md, where every registered candidate is killed and the number survives an independent recount. | X | [xchan-at37-offset-prereg.md](history/staging/xchan-at37-offset-prereg.md) |\n| `Q-xchan-at37-score` | ANSWERED | What does the @37 census say about the sealed X-channel offset pre-registration? | The census measured 1 - J = 0.020823, below every registered prediction, so scored exactly as registered every one of the seven candidates dies at \\|z\\| = 104 to 122 and the survivor set is EMPTY, an outcome the sealed prereg has no consequence clause for; the number itself survives an independent recount on a different marking scheme, and the offset question is replaced by a larger one. | X | [xchan-at37-score.md](history/staging/xchan-at37-score.md) |\n| `Q-xchannel-at23` | PARTIAL | What is the fifth point of the X-channel statistic, at @23, and does the monotone rise hold? | The fifth point is +0.2658 at @23, the rise holds at five levels and is larger than the four-point trend predicted, and the deficit has migrated into m >= 3, whose share of the X-gap runs 10.5, 23.5 and 81.1 per cent at @17, @19 and @23; the constant itself is not derived here. | X | [xchannel-at23.md](history/staging/xchannel-at23.md) |\n| `Q-xchannel-closedform` | PARTIAL | Does the zero-parameter closed form for the joint deficit survive its two blind tests at @29 and @31? | 1 - J = 4S2 is the only one of the three pre-registered laws left standing and it is not exact: a clean hit at @29, z = -0.90, and 4.93 sigma low at @31 where sigma_J falls to 0.000024, with the two rivals dying at 6.35 and 20.81 sigma; both blind levels agree on a stable relative offset of about half a percent, -0.41% and -0.48%, whose absolute residuals match to 2%. The @31 figure is quoted at the run's own sigma_J = 0.000024, and audit #85 (accepted, verified, its also_fix unapplied) records that the note's section 7 calls that sigma inherited rather than derived: under the slot-clustered sigma calibrated in `item-x-offset.md` section 4 on seven known-truth controls the same residual reads z = -2.32, inside the pre-registration's own HIT band, so this row's \"one detection at 4.93 sigma\" is zero detections at the pre-registration's threshold; the adjacent `Q-xchannel-offset` row quotes -0.43 and -2.32 for this same law at the same two levels, a factor 2.09 and 2.13 apart. | X | [xchan-at29.md](history/staging/xchan-at29.md) |\n| `Q-xchannel-offset` | PARTIAL | What produces the ~3.8 law's -0.45% offset, and which candidate correction survives? | The best available description is a finite-level correction of S3's size and sign - 1 - J = 4S2 - S3 scores z = +0.04 at @29 and -0.41 at @31 where the standing law scores -0.43 and -2.32 - but it is POST HOC; two candidate corrections and one shape are refuted on existing data, the sigma behind the @31 detection is calibrated on seven known-truth controls, and the deciding blind test at @37 has RUN: `Q-xchan-at37-score` (ANSWERED, 2026-08-21) measures 1 - J = 0.020823 there, below every registered prediction, so scored exactly as registered every one of the seven candidates dies at |z| = 104 to 122 with a 13.4*S3 shortfall and an EMPTY survivor set, and the finite-level-correction family is CLOSED as the offset's description, this row's 4S2 - S3 candidate with it; the @37 note's proposed TODO item X update is recorded there as \"report only, not applied\". | X | [item-x-offset.md](history/staging/item-x-offset.md) |\n| `Q-xchannel-triples` | ANSWERED | Does the joint deficit deepen or decay as the mixed super-W triple census is extended past @17? | It decays: J = 0.9025, 0.9599, 0.9659 at @17, @19, @23, so the deficit runs 9.7%, 4.0%, 3.4%, both new points land inside the pre-registered DECAYING FASTER THAN FORCED band, and the DEEPENING instinct is refuted by 3.50 sigma between @17 and @19 and a further 1.68 sigma to @23. | X | [xchannel-triples.md](history/staging/xchannel-triples.md) |\n| `Q-y2-recompute` | ANSWERED | Does the sixteen-level Y2 ladder reproduce on the corrected instrument, and does Y2/x^2 fall? | The ladder reproduces exactly, sixteen of sixteen from sixteen independent processes, digit for digit against the published 2026-08-18 ladder, and Y2/x^2 falls at 15 of the 15 consecutive steps; this does NOT show that G2(x#)/x^2 falls, since the shortfall Y2/(G2-1) is unmeasured above x = 79. | 1d | [y2-recompute.md](history/staging/y2-recompute.md) |\n| `Q-z2-state` | PARTIAL | What is known about Z2, the largest gap between twin openers inside the zone (p, p'^2), and at what calibration? | Draft consolidation only, computes nothing new: the zone statement is an interval statement, the parity obstruction applies to it in full, and no proven statement about Z2 survives past the zone's own width. | none | [z2-state-draft-0829.md](history/staging/z2-state-draft-0829.md) |\n| `Q-z3-immune` | CLOSED | Do the square anchor's immune offset classes (stretch-01 §4) carry a survivor surplus beyond the redistribution-only guardrail, do they attract the floor's first survivor, and does restricting the transplanted caps to them lower K*? | No, on all three. The preregistered surplus test is VOID on its own matched control (a 0.6% ratio-of-means bias present with and without the quadratic point); the preregistered placement test HOLDS at Z = +0.49; the design-free class test reads the guardrail's r/(r-2) over 3.4 million twin openers with a largest per-prime deviation of 0.11% and per-prime Z in -0.48..+1.66; and the restricted certificate, while lowering K* at 61% of anchors, has no certificate at all at 93 of 1219 and is beaten head to head by the QR-BLIND restriction that certifies 1219 of 1219. The refinement is closed. | Z3 (retired) | [attack-z3-immune-01.md](history/staging/attack-z3-immune-01.md) |\n| `Q-zone-postulate` | OPEN | Is every zone occupied by a twin pair, and what is proven toward it? | OPEN: the weak form is PROVEN equivalent to the twin prime conjecture (both directions, elementary), the zone's supply is MEASURED to grow like p^2/ln^2 p with margin about p and verified by segmented sieve to 1e11, and the obstruction is named; none of the four attack routes closes it. | none | [ZONE-POSTULATE.md](ZONE-POSTULATE.md) |\n| `Q-zone-tail` | PARTIAL | What is the tail field of the zone (p, p'^2) - its law against ln^2, its worst case, its share of the R0 width, and is it the frozen-sieve last survivor below p'^2 at the rate the head is the frozen sqrt(p)-level first survivor above p? | The field exists now, 1,225 zones, and nothing in it is derived. The tail is the head's own renewal functional read at height p'^2 rather than p: c_tail = 0.7771 ln^2(p'^2) at [3163,1e4) against the head's 0.6693 ln^2 p on the same zones and the same estimator, both referenced to HL's 0.7574, and the apparent factor of four is the unit ln^2(p'^2) = 4 ln^2 p and nothing else. The band drift does not settle (0.7518, 0.7288, 0.7771; the pre-registered 0.03 step is missed), so this is band composition on two decades and not a measured convergence. The frozen-sieve mirror is 100% at the zone's own level p, but that is the Zone Restriction Lemma restated and not a rate comparable to the head's 96.65%; the non-vacuous mirror is one fold back, where the tail survives at 98.69% with the residual priced exactly (88 extra lpf(n) = p survivors, 18.18% of them biting), and the head's own frozen level sqrt(p) identifies the tail in only 22.86%, refuting this note's own pre-registered [0,5] zones. The renewal surplus t/R = 1.0619 at B4 is NOT RESOLVED, its bootstrap [0.9937, 1.1321] containing 1, and an exhaustive class-matched origin null over 5,962,057 origins gives the same 1.0619; three matched non-square endpoints see no difference at all, sign tests z = 0.14, -0.54, 0.66 with six paired intervals all containing zero, so nothing here is a property of the endpoint being a prime square. R0 is asserted at all 1,225 zones and the shares are 1.82% head, 89.72% Z2, 8.46% tail at B4 with the tail's share FALLING, so filling the tail row moves the decomposition's difficulty nowhere: Z2 was the obstruction before and is more of it after. | Z4 | [zone-tail-01.md](history/staging/zone-tail-01.md) |\n| `Q-zone-tail-02` | ANSWERED | On the full zonegap-01 range to 1e11, what is the per-zone tail law in both units, is the renewal surplus t/R real at 27,292 zones, and do the head and the tail of a zone have a joint law? | Resolved DOWN and nothing is derived: zone-tail-01's unresolved t/R = 1.0619 at 782 zones reads 1.0298 on a height-matched comparator at 17,700 zones (bootstrap [1.0156, 1.0443], excluding 1) and 1.0157 against the exactly matched discrete class null (bootstrap [1.0013, 1.0308], clearing 1 by 0.0013), so part of the original figure was the comparator and the pre-registered rule returns MATCH by two parts in ten thousand; c_local = 0.7522 at the top band with bootstrap [0.7410, 0.7630] now CONTAINING HL's 0.7574 and the band drift inside the registered 0.03 per step for the first time but still non-monotone (0.7518, 0.7664, 0.7416, 0.7522), so no convergence is measured; global worst tail/ln^2(p'^2) = 6.840 at p = 15107 with max/mean no longer flat (4.40, 5.36, 8.59, 8.43); Q5 is answered in the negative, head and tail showing no dependence at 27,292 zones (detrended r = -0.0009 against null s.e. 0.0061, permutation p = 0.8920, variance ratio 0.9990, no KS rejection, no chi-square near threshold), so the R0 pieces are a product structure to first and second moments and in the sum's whole law while independence is NOT proven and the simultaneous-extremes regime is untouched; a candidate rough-origin mechanism for the surplus is HEURISTIC and does not fit the head's own excess band by band; the head field's effective sample size is measured at 1,910 distinct a_first among 17,700 top-band zones, so this note's own head interval is priced over draws the field does not hold, a cluster bootstrap over distinct a_first (producer SEC I(a), seed 613009117) measuring the inflation at 4.10 rather than the 3.04 the redundancy alone implies and moving the top-band interval to [0.6848, 0.7692], still inside the registered [0.68, 0.78] (correction originating with redteam-0829-measure-b.md section 1c C15); the scope of that defect is that one interval and not the corpus, since destroyer-census-01.md publishes no bootstrap, head-residual-factor.js bootstraps over gaps, zonegap-01.js has none and zone-tail-01.js bootstraps only tail quantities; ten of ten predictions hit, which is a criticism of the pre-registration; and the R0 shares read 1.40% / 92.80% / 5.81% with the tail's share still falling, so Z2 is more of the obstruction after this pass than before it. | Z4 | [zone-tail-02-0829.md](history/staging/zone-tail-02-0829.md) |\n| `Q-zonegap-01-prereg` | SUPERSEDED | What is Z2(p) in the unswept half-decade p in [1e5, 316228)? | Pre-registration only, committed before the X = 1e11 sweep ran: five predictions are sealed on the per-zone distributional structure the published record ladders do NOT determine, with the records excluded from scoring and used as custody instead; the scoring document is the stage-2 embedded output, in zonegap-01.md. | none | [zonegap-01-prereg.md](history/staging/zonegap-01-prereg.md) |\n| `Q-zonegap-03-prereg` | OPEN | What is Z2(p) in the unswept decade X in (1e11, 1e12]? | Pre-registration only, sealed before any sweep past X = 1e11 exists: five theorem-grade predictions at sigma = 0 (resting on Z2(p) = env(p) identically, a two-line theorem conditional on the adopted ladder being the true running max) and five blind statistical ones with sigma models and kill rules, scored by the stage-3 embedded output with misses stated first. | none | [zonegap-03-prereg.md](history/staging/zonegap-03-prereg.md) |\n| `Q-zonegap-03-score` | ANSWERED | Do the ten predictions sealed in zonegap-03-prereg.md score against the stage-3 sweep at X = 1e12? | All ten sealed rows score HIT and none miss, after a second pass added a band-edge argument to zonegap-01.js and re-ran the decade so the two rows that named an unprinted band could be scored; the four Group T hits are a custody promotion that follows from CUSTODY 1 and 3 passing, one sub-clause of T5 (the full-decade sd) stays unprinted, and the sweep still needs a DERIVED engine because zonegap-01.js's inlined 41-record ladder makes it exit at 1e12. | Z7 | [zonegap-03-score.md](history/staging/zonegap-03-score.md) |\n| `Q-zonegap-Z2` | PARTIAL | What is Z2(p), the largest twin-pair gap inside the zone (p, p′^2), measured at every zone to 1e11? | MEASURED to 1e11 over 27,292 zones: every zone holds at least two twin pairs, Z2 is the record envelope wearing zone coordinates, Z2(p) <= G2(p#) is a one-line theorem with equality only to p = 7, and the zone-local margin runs away like p^2/(3.9 ln^3 p); the pre-registration scored 2 of 5. | none | [zonegap-01.md](history/staging/zonegap-01.md) |\n| `Q-zonegap-model` | ANSWERED | What is Z2 once decomposed, and does the per-zone field carry anything the record ladder does not? | Z2(p) = env(p) with D(p) identically zero (PROVEN conditional on the adopted ladder, VERIFIED 204 ways), so the per-zone field is the record ladder wearing zone coordinates and the u bimodality is closed; the blind tests for the [1e11, 1e12] decade are sealed and the sweep itself is not run. | none | [zonegap-03-model.md](history/staging/zonegap-03-model.md) |\n| `Q-zonegap-prior-art` | ANSWERED | Who owns Z2(p), the maximal twin-prime gap below a quadratic cutoff, in the owning conventions? | Kourbatov (2013) and Kourbatov-Wolf (2019) own maximal twin-prime gaps, with heuristics and records and nothing proven; the zone form (p, p′^2) has five neighbours and no owner, and the per-zone quadratic-cutoff object appears absent on channels calibrated in the same session. | none | [zonegap-prior-art.md](history/staging/zonegap-prior-art.md) |\n| `Q-zonegap-reduction` | PARTIAL | What must be proven for Z2, what already is, and what does the proven head machinery give the zone? | The Zone Restriction Lemma is proven and verified at five levels - the in-zone twin slots of T_p ARE the in-zone twin primes, so Z2(p) IS the whole-tile gap object restricted to the head window - the three-slack chain is graded honestly, the transfer inventory is listed item by item, and the deep end is proven an onset desert over 24 levels; section 5 lists what was not reached. | Z (retired) | [zonegap-02-reduction.md](history/staging/zonegap-02-reduction.md) |\n| `Q-zonegap-witnesses` | ANSWERED | Do independent second witnesses confirm the twin-gap record table, and does the due OEIS flip-search turn anything up? | Both witnesses AGREE at 82/82 rows with no extension and no conflict and the lesser-to-lesser convention verified rather than assumed; the OEIS check returns one loud hit, the per-zone pair-count sequence being A273257 one index over, while the gap objects stay absent, and one literature item, Korevaar 2012, is still walled and read only by arXiv proxy. | none | [zonegap-witnesses.md](history/staging/zonegap-witnesses.md) |\n\n## 3. Unindexed notes (0): title only, until a block is added\n\n*(Legacy notes written before the ledger existed. A grep for the object still\nlands here. Add a block when a note is next touched; `node research/qc.js\nledger-backlog` lists them.)*\n\n", "mkpatch3852.py": "#!/usr/bin/env python3\n# Job #609: build the audit patch for the four source-stale rows of\n# research/QUESTIONS.md (38, 39, 47, 48). Each corrected row text follows the\n# served source that supersedes it, quoted in the return. Every old string must\n# occur exactly twice (section 1 and section 2 of the generated index) and must\n# not occur inside the new string.\nimport difflib, hashlib, io, sys\n\nP = \"QUESTIONS.md\"\nsrc = io.open(P, encoding=\"utf-8\", newline=\"\").read().replace(\"\\r\\n\", \"\\n\")\n\n\ndef slice_between(a, b):\n i = src.find(a)\n if i < 0:\n raise SystemExit(\"start anchor not found: %r\" % a[:60])\n j = src.find(b, i)\n if j < 0:\n raise SystemExit(\"end anchor not found: %r\" % b[:60])\n return src[i:j + len(b)]\n\n\nEDITS = []\n\n# --- row 38, Q-rec-cheapest-0830 -------------------------------------------------\nold38 = slice_between(\"the exit-chain count is DERIVED (held)\", \"under z^{u0})\")\nnew38 = (\n \"the exit-chain count is DERIVED (held) to grow like z^{16s/9 - o(1)} for \"\n \"s <= 3, the hypothesis the 2026-08-30 red team added to the threshold, and \"\n \"z^{2s - o(1)} in the limit, so REC(s, u0) is false for every u0 < 16s/9 = \"\n \"4.7088 at s -> 1+sqrt(e), the whole legal band (2, beta_2] included, and \"\n \"RML(alpha) with it, its falsity beyond z ~ 5.6e31; at every computable z it \"\n \"is a proof gap (Omega(z) exact to z = 73 reads 251, and the ledger's former \"\n \"comparator 1.05e6 is HM(u0), not z^{u0} = 7.188e7 at z = 73, so the gap is \"\n \"68.3x larger than the ledger stated, while a certified lower bound 8.86e17 \"\n \"at z = 5e5 sits a factor 1.2e6 under z^{u0})\"\n)\nEDITS.append((old38, new38))\n\n# --- row 39, Q-recon-0904-sifting-limit-floor ------------------------------------\nold39 = slice_between(\"the cap at 4.26645 is a METHOD ARTEFACT at rung MEASURED\",\n \"the band (2, 4.26645] is untouched by anything proven.\")\nnew39 = (\n \"the cap at 4.26645 as a METHOD ARTEFACT at rung MEASURED is REFUTED as \"\n \"stated (rider 2026-09-04, from `redteam-0904-sifting-limit.md`, on four \"\n \"independent grounds): the level-D LP fixes one omega-profile out of the \"\n \"class Ford defines by a one-sided inequality, and a legal profile inside the \"\n \"classical budget at x = 13 reads 5.0113 calibrated, above beta_2, firing the \"\n \"note's own pre-registered refutation; the ledger's \\\"3.3152 sits about 18 per \"\n \"cent below the DHR value\\\" is 22.3 per cent; and its \\\"passed, 0 of 15\\\" \"\n \"external test does not survive, so the calibrated sentence is that no barrier \"\n \"theorem exists, no axiom-only argument inside the band is exhibited, and the \"\n \"LP number is a finite-level measurement for one profile; the band \"\n \"(2, 4.26645] is untouched by anything proven.\"\n)\nEDITS.append((old39, new39))\n\n# --- row 47, Q-xchannel-closedform ----------------------------------------------\nold47 = \"match to 2%. |\"\nnew47 = (\n \"match to 2%. The @31 figure is quoted at the run's own sigma_J = 0.000024, \"\n \"and audit #85 (accepted, verified, its also_fix unapplied) records that the \"\n \"note's section 7 calls that sigma inherited rather than derived: under the \"\n \"slot-clustered sigma calibrated in `item-x-offset.md` section 4 on seven \"\n \"known-truth controls the same residual reads z = -2.32, inside the \"\n \"pre-registration's own HIT band, so this row's \\\"one detection at 4.93 \"\n \"sigma\\\" is zero detections at the pre-registration's threshold; the adjacent \"\n \"`Q-xchannel-offset` row quotes -0.43 and -2.32 for this same law at the same \"\n \"two levels, a factor 2.09 and 2.13 apart. |\"\n)\nEDITS.append((old47, new47))\n\n# --- row 48, Q-xchannel-offset --------------------------------------------------\nold48 = \"and the deciding blind test at @37 is pre-registered and committed alone.\"\nnew48 = (\n \"and the deciding blind test at @37 has RUN: `Q-xchan-at37-score` (ANSWERED, \"\n \"2026-08-21) measures 1 - J = 0.020823 there, below every registered \"\n \"prediction, so scored exactly as registered every one of the seven \"\n \"candidates dies at |z| = 104 to 122 with a 13.4*S3 shortfall and an EMPTY \"\n \"survivor set, and the finite-level-correction family is CLOSED as the \"\n \"offset's description, this row's 4S2 - S3 candidate with it; the @37 note's \"\n \"proposed TODO item X update is recorded there as \\\"report only, not applied\\\".\"\n)\nEDITS.append((old48, new48))\n\nnew_src = src\nfor old, new in EDITS:\n n = new_src.count(old)\n if n != 2:\n raise SystemExit(\"old string occurs %d times, expected 2: %r\" % (n, old[:70]))\n if old in new:\n raise SystemExit(\"new string still contains the old: %r\" % old[:70])\n new_src = new_src.replace(old, new)\n\nif new_src.count(\"\\n\") != src.count(\"\\n\"):\n raise SystemExit(\"line count changed\")\n\ndiff = difflib.unified_diff(src.split(\"\\n\"), new_src.split(\"\\n\"),\n fromfile=\"a/research/QUESTIONS.md\", tofile=\"b/research/QUESTIONS.md\",\n lineterm=\"\", n=2)\npatch = \"\\n\".join(diff) + \"\\n\"\nio.open(\"job609/questions-rows38-52.patch\", \"w\", encoding=\"utf-8\", newline=\"\").write(patch)\nio.open(\"job609/QUESTIONS.corrected.md\", \"w\", encoding=\"utf-8\", newline=\"\").write(new_src)\n\nchanged = sum(1 for ln in patch.split(\"\\n\") if ln.startswith((\"+\", \"-\")) and not ln.startswith((\"+++\", \"---\")))\nprint(\"edits:\", len(EDITS), \" changed lines in patch:\", changed)\nprint(\"served QUESTIONS.md sha256(raw):\", hashlib.sha256(io.open(P, \"rb\").read()).hexdigest())\nprint(\"patched QUESTIONS.md sha256(LF):\", hashlib.sha256(new_src.encode()).hexdigest())\nfor f in (\"job609/questions-rows38-52.patch\", \"job609/QUESTIONS.corrected.md\"):\n raw = io.open(f, \"rb\").read()\n print(f, len(raw), \"sha256(raw):\", hashlib.sha256(raw).hexdigest())\n"}, "docs": {"research/history/staging/attack-0829n-rml-proof.md": "# The proof attempt on RML: the chain from the proven mean square to an exponent below β₂, and the one arrow that does not close\n\n\n\n> **RIDER 2026-08-30 (orchestrator, from `redteam-0830-rml.md`: 12 claims\n> stand, 14 fail at sentence level, none inflating any result — the\n> load-bearing one errs in the CONSERVATIVE direction).** PROVEN under\n> independent code: the whole CAP lemma of §4.1, including the sine-sum step\n> this note checked only numerically (Σ 1/sin(πa/e) ≤ e·H_{⌊(e−1)/2⌋} + 1 ≤\n> e(ln e + 1), violated at no e ≤ 20000), the slope 2.7660 ± 0.2120 with its\n> 1.10 and 5.82 se placements, and the reduction. WHAT FAILS: (i) §2's\n> `ms_m`/`ask` column is log_z √(B·H) with B the PROVEN Lemma V constant, i.e.\n> the exponent of the proven BOUND on the rms, not of the rms ⟨R_H²⟩^{1/2}\n> that REC names; the two part by 36.53× to 200.45× because ⟨R_H²⟩ saturates\n> once H passes the correlation length while B·H grows linearly, so §5's\n> reading \"at u₀ = 3.0 and 3.5 ask does not exceed gauss\" REVERSES at every\n> row when both sides are measured against the same mean square; (ii) \"gauss\n> → 0\" and \"sup/rms ≤ √(2 lnW)·O(1) = z^{o(1)}\" are both wrong (the limit is\n> 1/2, measured 0.5645 at z = 1e7) and contradict this note's own §0; (iii)\n> \"the recovery exponent … 0.99–1.29\" quotes the `truth` column, whose real\n> value is 0.56–0.67; (iv) \"REC would be moot rather than false\" should read\n> false; (v) \"u_cap → 2s\" and \"n ~ z^{2s}/polylog\" are asserted, not proven\n> (`verify-0830-usup-convention.md` §6's \"OPEN in both directions\" is the\n> better-calibrated reading); (vi) the 1.509-octave correction is still\n> unapplied in three places. NEW: u_cap crosses β₂ at z = 53, reaches 4.4891\n> at z = 79, and passes u_triv from z = 53.\n\n*(2026-08-29 night wave, staging. Producer:\n`research/history/staging/attack-0829n-rml-proof.js`, embedded via\n`qc/embed.js`. No existing file edited, no git command run. Every number\nbelow is either in the producer's OUTPUT block or a quotation of an embedded\nartifact cited by file and section. Calibration per claim: PROVEN, MEASURED,\nHEURISTIC, OPEN, REFUTED.)*\n\n## 0. What is open, first\n\n- **RML(α) is open at every α, and so is every form F1–F4. Nothing here\n lowers any exponent.** The deliverable is a location, not a theorem: the\n chain from Lemma V's PROVEN mean-square bound to an exponent below\n β₂ = 4.26645 has one open arrow, written down in §3 with its quantifiers,\n and none of the five attempts in §4 closes it.\n- **The open arrow is Lemma V's sup form (F2 of `rho-maximal-law.md` §1)\n with the proven mean square factored out.** It is not a new statement. What\n is new is the accounting: which part of F2 is proven (the rms), what the\n remainder must supply (a recovery of the sup from the rms by a factor\n z^{u₀/2−ε}), what every proven recovery actually costs (either the\n position-union price e^{θ(z)/q} or the term-count cap z^{2s+o(1)}), and\n what the truth measures (the sup exceeds the rms by about √(4θ(z)), i.e.\n z^{1/2+o(1)} if the Gaussian form persists).\n- **Finite-z sufficiency is calibration, not a result.** At z ≤ 37 even the\n term-count cap reads below β₂ (u_cap 3.4103 → 4.0593, OUTPUT S2), which is\n the Bonferroni-x ≤ 227 phenomenon of `rho-maximal-law.md` reading 4 at yet\n another instrument. No row of any table below is evidence for an asymptotic\n statement; the rows are there to show the arrow is true where it can be\n seen, and by how much.\n- **Asymptotically the arrow is undecided in both directions.** The truth's\n measured slope, 2.7660 ± 0.2120 on seven points spanning less than one\n octave of z (OUTPUT S5), leaves 1.23 exponents of room under u₀ = 4 and\n 0.23 under u₀ = 3; the corpus's standing observation that local slopes here\n rise with z (`rho-maximal-law.md` reading 6; `rho-exact-z31-01.md` §0)\n applies. The u₀ = 3 form could be false; the u₀ = 4 form is 5.8 se clear on\n a fit that cannot refute an asymptotic statement.\n- **What this note did not do.** It ran no new sup over positions (the exact\n points at z = 31, 37 were bought on 2026-08-29 by `rho-exact-z31-01.md`),\n it did not price the smooth-profile form again (CLOSED, `REFUTED.md`\n \"manufacturing the smooth profile\"), and it did not touch ⟨ρ̃⁴⟩, the named\n stopping object of `attack-rhoms-01.md` §3.\n\n## 1. The objects and the direction of every inequality (wrong-direction guard)\n\nFix s = 3.0 (the measured family; any s > 1 + √e = 2.6487 gives M > 0), a\nprime z, W = P(z), D = z^s, the divisor-pair lattice (block, d₁, d₂) with\nq = [d₁, d₂], w = ε μ(d₁)μ(d₂) ∈ {±1}, c the CRT class, n the term count and\nM = Σ w/q > 0 (`rho-maximal-law.md` §1). The potential ρ(y) = Σ_j w_j\nψ((y − c_j)/q_j), the centred ρ̃ = ρ + M/2, and for a window H the certificate\n\n T(x) = H·M + R_H(x), R_H(x) = ρ(x) − ρ(x + H),\n\nwith T(x) ≤ #{twin-admissible r ∈ (x, x+H]} pointwise (Brüdern–Fouvry,\nPROVEN, `rho-maximal-law.md` §2 (iv)). The consumer is min_x T(x) ≥ 1, which\ngives G₂(z#) ≤ H. Every inequality this note uses or asks for, with its\ndirection and why it is legal under `attack-wrongdirection-audit.md`:\n\n| # | inequality | direction | rung | legality |\n|---|---|---|---|---|\n| I1 | ⟨R_H²⟩ ≤ B(z,s)·H, B ≤ 9A²(E−1) = O(ln⁸z) | upper bound on a period mean | PROVEN (`lemmaV-parseval.js` L1–L5; `redteam-0829-theorem1.md`) | no position quantifier; gives only almost-all statements at any H, so it carries no TPC content even at H < z² |\n| I2 | sup_x \\|R_H(x)\\| ≤ H·M − 1 at H = z^{u₀}, ONE fixed u₀ ∈ (2, β₂) | upper bound on the remainder, the direction that bounds G₂ from above | OPEN (= F2 at (s, u₀)) | legal by audit row 10: the band (2, β₂] is exhibited legal, α = β₂ being PROVEN with TPC open. TPC-strength if u₀ ≤ 2 (Gap Reformulation, constant < 1 at exponent 2) or if stated uniformly in u down to 2 (Axis A). This note fixes u₀ and never lets it move |\n| I3 | REC(s,u₀): sup_x \\|R_H(x)\\| ≤ z^{u₀/2−ε}·⟨R_H²⟩^{1/2} at H = z^{u₀} | upper bound on a sup by a multiple of an rms | OPEN, the arrow | implies I2 through I1 and M ≍ 1/ln²z; same legality as I2, same trap at u₀ = 2 (there it would read sup ≤ z^{2−ε}·polylog < z²/ln²z, i.e. o(z²), which is TPC) |\n| I4 | sup\\|R_H\\| ≤ Σ_e Vabs(e)·e·(ln e + 1) =: CAP(z) for every H | upper bound, absolute values across moduli | PROVEN here (§4.1) | harmless: delivers exponent 2s + o(1) ≥ 2(1+√e) = 5.2974 |\n| I5 | M ≥ c(s)/ln²z at s > 1+√e | lower bound on the main term | standard, CITED; MEASURED M·ln²z = 0.3359–0.3772 at z = 13..37 (OUTPUT S0) | no twin content |\n| I6 | the floor: no correct law delivers H below nP(z); th(nP) ≥ 2.0106 at z = 47 | lower bound on any deliverable window | MEASURED (`rho-maximal-law.md` §2, rider 2) | the reverse direction; if it ever exceeded β₂ it would kill the item, and it is undecided asymptotically |\n\nTwo things the table makes explicit. I1 is the strongest possible statement\nabout the remainder on average: at every level D and every H the mean square\nis √H·polylog, square-root cancellation with no level restriction, and it is\nproven. The consumer needs something far weaker in size, H/ln²z, but\npointwise. The whole distance between I1 and I2 is the quantifier over x,\nwhich the corpus has priced before (`sift-limit-attack.md` §7e: \"the price is\nexactly the quantifier\"); what §3 does is write that price as one inequality\nwith its exponent.\n\n## 2. The chain, each arrow graded\n\nThe vector sieve consumes one statement, min_x T(x) ≥ 1 at some H = z^{u₀}.\nRead from the proven end:\n\n```\n (C1) <= B(z,s) H PROVEN L1-L5 (lemmaV-parseval.js); every H, z, s\n (C2) B(z,s) <= 9 A(z)^2 (E(z)-1) = O(ln^8 z) PROVEN redteam-0829-theorem1.md; no hypothesis\n => rms(R_H) <= z^{u0/2} (ln z)^4 O(1) PROVEN composite of C1, C2 at H = z^{u0}\n (C3) sup_x |R_H(x)| <= z^{u0/2 - eps} rms OPEN REC(s,u0), the arrow, sec.3\n (C4) M >= c(s)/ln^2 z (s > 1 + sqrt e) CITED standard linear sieve; MEASURED 0.3359-0.3772 (S0)\n => sup_x |R_H(x)| <= H M - 1, z >= z0 PROVEN from C3 + C2 + C4 (the polylogs are absorbed by z^eps)\n (C5) T(x) = H M + R_H(x) PROVEN exact identity\n (C6) T(x) <= #{admissible r in (x, x+H]} PROVEN Bruedern-Fouvry pointwise, pilot-verified\n => G2(z#) <= H = z^{u0}, z >= z0 PROVEN from C4-C6: exponent u0 < beta_2\n```\n\nBeside the open arrow stand two proven arrows that reach the consumer\nwithout it, and both land above β₂:\n\n```\n (C3-triv) sup|R_H| <= 2 sup|rho~| <= n + M PROVEN triangle inequality over the lattice\n => exponent u_triv -> 2s + o(1) at s = 3: 6; at s -> 1+sqrt e: 5.2974\n (C3-cap) sup|R_H| <= CAP(z) = sum_e Vabs(e) e (ln e + 1) PROVEN sec.4.1, this note\n => exponent u_cap -> 2s + o(1) same asymptote; 3.4103 -> 4.0593 at z = 13..37 (S2)\n (C3-cheb) #{x : |R_H(x)| >= HM-1} <= B H W/(HM-1)^2 PROVEN Chebyshev off C1\n => almost-all exponent 0, sup exponent none attack-beta2-01 item 2; rho2-analytic-bound.md sec.4\n```\n\nAt numbers, s = 3.0, from OUTPUT S1 (exponents base z; rows with H > W\ndropped):\n\n| u₀ | z | need = log_z(HM−1) | ms_m = log_z √(B_meas H) | ask = need − ms_m | union = θ(z)/(2 ln z) | truth = log_z(2 sup\\|ρ̃\\|) |\n|---|---|---|---|---|---|---|\n| 4.00 | 19 | 2.9033 | 2.0613 | 0.8420 | 2.2319 | 0.9873 |\n| 4.00 | 29 | 2.9763 | 2.0568 | 0.9195 | 2.8544 | 1.0626 |\n| 4.00 | 37 | 3.0104 | 2.0551 | 0.9553 | 3.6036 | 1.2874 |\n| 3.00 | 19 | 1.9021 | 1.5613 | 0.3408 | 2.2319 | 0.9873 |\n| 3.00 | 37 | 2.0102 | 1.5551 | 0.4551 | 3.6036 | 1.2874 |\n| 4.25 | 37 | 3.2604 | 2.1801 | 1.0803 | 3.6036 | 1.2874 |\n\nRead across a row: the consumer at (3.0, 4.0), z = 37 needs sup|R_H| under\nz^{3.0104}; the proven mean square hands over an rms of z^{2.0551}; so the\narrow must recover the sup from the rms within a factor z^{0.9553}\n(asymptotically z^{u₀/2 − o(1)} = z^{2−o(1)}, the finite-z figure being\nsmaller because M and the polylogs bite); the only fully proven recovery,\na union over the W positions, charges z^{3.6036} and rises like z/(2 ln z);\nand the truth, certified from the exact sup|ρ̃|, sits at z^{1.2874}. The\narrow is the inequality truth ≤ ask, and at every measured level it holds\nby a wide margin (§5); the union price exceeds the ask at every level from\nz = 19 on and diverges.\n\n**Which arrow is \"the\" open one.** C3 is the only arrow in the first block\nthat is not PROVEN or CITED-standard. C4's citation is standard linear-sieve\nasymptotics at s > 1+√e and is not in doubt; C6 is Brüdern–Fouvry's pointwise\ninequality, verified. So (a) of the brief is answered: one open arrow, C3,\nand everything else in the chain is a theorem.\n\n## 3. The open arrow, isolated, with its quantifiers and its payoff\n\n**REC(s, u₀).** Fix s > 1 + √e and u₀ ∈ (2, β₂). There exist ε > 0 and z₀\nsuch that for every prime z ≥ z₀, with D = z^s, H = ⌊z^{u₀}⌋, W = P(z), the\nsigned sawtooth remainder of the level-D vector-sieve lattice satisfies\n\n sup_{x ∈ ℤ/W} |R_H(x)| ≤ z^{u₀/2 − ε} · ⟨R_H²⟩^{1/2},\n\nthe mean on the right taken over one full period. Quantifier discipline: the\nsup is over ALL W positions; s and u₀ are fixed before z; ε may depend on\n(s, u₀) but not on z; u₀ is never a function of z and is never sent to 2.\n\n**Equivalent forms, so the arrow can be recognised in other clothing.**\n\n- In sieve currency: a level of distribution for the signed two-class\n Rosser comb in short intervals, uniformly in the interval's position. The\n joint level is D² = z^{2s} = H^{θ_total} with θ_total = 2s/u₀; at (3.0, 4.0)\n that is θ_total = 1.5, and the break-even the corpus has already priced is\n θ_total > 1.2090 for Brüdern–Fouvry's asymmetric split\n (`sift-limit-attack.md` §7b) and 5.297442541400/β₂ = 1.241651 for the\n symmetric lattice used here (§7b (1b)). REC is that level statement with\n the remainder's size measured against its own rms rather than against H.\n- In the F-ladder of `rho-maximal-law.md` §1: REC(s, u₀) is F2 at window\n exponent u₀ with the mean square factored out. F2 ⟸ REC, and F3 = RML(α)\n with α < u₀ ⟹ REC (through |R_H| ≤ 2 sup|ρ̃|). F4 (the Gaussian form,\n λ = 0) ⟹ REC with room to spare: it supplies sup/rms ≤ √(2 lnW)·O(1) =\n z^{o(1)}, against an allowance of z^{u₀/2 − ε}.\n- The weakest law-shaped form beneath it is per-position in H: ∀x ∃H ≤ z^{u₀}\n with T_H(x) ≥ 1. It is implied by REC and it is exactly \"every window of\n length z^{u₀} contains a certified survivor\"; no instrument in the corpus\n distinguishes it from F1, and it is listed so the ladder has its bottom.\n\n**What its truth gives, quantified.** REC(s, u₀) ⟹ G₂(z#) ≤ z^{u₀} for all\nz ≥ z₀ (chain C4–C6), an unconditional two-class exponent u₀. At u₀ = 4.0 the\nmovement below β₂ is 0.266 of exponent; at u₀ = 4.25 it is 0.016; at\nu₀ = 3.0 it is 1.266 [ARITHMETIC: β₂ − u₀ on the cited β₂], and the u₀ = 3.0\nform is the one the data leave least room for (§5). Every u₀ here is above the measured floor th(nP) ≥ 2.0106\n(z = 47, I6), so none is excluded by the floor at any measured level.\n\n**What it costs, in the corpus's own currencies.** Against the trivial\naccounting (C3-triv, C3-cap) REC asks for a saving of z^{2s − u₀ − o(1)} over\nthe triangle inequality, uniformly in x: z^{2.0} at (3.0, 4.0), z^{1.75} at\n(3.0, 4.25), and z^{1.031} at the cheapest legal point (s → 1+√e, u₀ → β₂)\n[ARITHMETIC: 2s − u₀, and 5.2974 − 4.2665 on the cited constants], which is\nthe 2(1+√e) − β₂ gap the corpus has carried since `sift-limit-attack.md`\n§4.5. Against the proven mean square it asks for a\nrecovery exponent below u₀/2, where the proven recoveries cost θ(z)/(2k ln z)\nat moment order k (`attack-rhoms-01.md` §3), i.e. diverge. The measured truth\nof the recovery exponent is log_z(2 sup|ρ̃|/rms): about 0.99–1.29 in the\ntable of §2, and if the Gaussian form persists it is 1/2 + o(1).\n\n**Kill checks, run.** (i) Is REC on a REFUTED row? No. The rows nearest it\nare \"Lemma V's mean-square form as the missing factor\" (CLOSED because B was\nnever binding, which is the statement C1–C2 are proven and irrelevant to C3),\n\"u_sup as an unconditional worst-position bound\" (CLOSED; C3-cap is that route\nwith its asymptote named, not the arrow), \"the ℓ¹ → ℓ²√log conversion\"\n(CLOSED because that conversion IS the sharp maximal law; REC asks for\nℓ² × z^{u₀/2}, not ℓ² × √log), \"generic chaining\" (CLOSED; not retried), and\n\"the Kowalski–Michel–Sawin branch\" and \"manufacturing the smooth profile\"\n(CLOSED; §4.5 says why they are not REC). (ii) Is REC TPC-strength? No, at\nfixed u₀ > 2 (§1, I3). (iii) Is it already listed ANSWERED? No: QUESTIONS.md\ncarries Q-rho-maximal-law as PARTIAL with the law itself open, and this note\nis the proof attempt that item calls for, not a re-run of its measurements.\n\n## 4. The attempt, and where it dies\n\nFive routes were tried on REC. One produced a small theorem (4.1, an\nasymptote for a closed route); none produced the arrow. Each is recorded\nwith the exact place it stops.\n\n### 4.1 Term-by-term accounting is capped at 2s + o(1), and the cap is the trivial-level exponent\n\n**Lemma (PROVEN here).** For every H, sup_x |R_H(x)| ≤ CAP(z) :=\nΣ_{e | P(z), e > 1} Vabs(e)·e·(ln e + 1), and CAP(z) ≤ z^{2s}·(2s ln z + 1)·\n6·Π_{p β₂ at every admissible\ns; the divergence disjunct is now refuted at fixed s.\n\n### 4.2 Position-union routes, including hypercontractivity, are dominated by 4.1\n\nChebyshev and the moment ladder are on record: sup ≤ W^{1/2k}·‖R_H‖_{2k}\npays e^{θ(z)/2k} and needs k ≍ z/(1.53 ln z) (`attack-rhoms-01.md` §3;\n`rho2-analytic-bound.md` §4). The one route not on record is the product\nstructure of the positions: x ∈ ℤ/W ≅ Π_{p 2s + log_z\n(#moduli): strictly worse than 4.1. Rung: the negative is elementary; no\nhypercontractive theorem was cited because its best case is already\ndominated. **Where it dies:** the atom √p, i.e. one residue class has no\ncancellation by itself; the saving must come from the signs w_j across\nterms, which no per-term or per-coordinate bound sees.\n\n### 4.3 Window smoothing and the P(y) lever move polylogs, not the exponent\n\nAveraging T_H(x) over H ∈ [H₀, 2H₀) is legitimate for the consumer (an\naverage ≥ 1 forces some T_H(x) ≥ 1, hence a survivor in (x, x + 2H₀]) and it\nreplaces |S_H(a/e)| ≤ 1/(2‖a/e‖) by a Fejér-type kernel ≤ c/(H₀‖a/e‖²),\nwhich cuts the per-modulus ℓ¹ mass from e ln e to about e²/H₀ for e < H₀\nand does nothing for e > H₀. Choosing H ≡ 0 mod P(y) kills every modulus\nall of whose primes are < y exactly (V1). Measured: at u₀ = 4 the moduli\nabove the window carry 0–27.26% of CAP over z = 13..37 (S3), so at\ncomputable z the smoothing would remove most of the ℓ¹ mass, and the P(y)\nlever at y = ⌈4 ln z⌉ removes 31.64% of CAP at z = 13 falling to 1.713% at\nz = 37 (S4). Neither touches the asymptote: **the part of CAP below\nz^{u₀ − δ} is at most z^{u₀ − δ}·polylog (PROVEN, from the second\ninequality of 4.1 applied to e ≤ z^{u₀ − δ}), so it is below HM − 1 for\nlarge z at any fixed δ > 0, and REC needs proving only for the moduli in\n(z^{u₀ − δ}, z^{2s}]**. At (u₀, δ) = (4, 0.5) that part reads 0.3948, 0.3884,\n0.5067, 0.5346 of HM − 1 at z = 19, 29, 31, 37, already below 1 but rising\n(the polylog dominates the z^{−δ} at these z), and the remaining moduli carry\n24.45–56.84% of CAP (S3). **Where it dies:** the moduli above the window\nare exactly the terms whose class hits the window at most once, so the\nreduced arrow is a bound on the signed count of large-modulus hits,\nΣ_{r ∈ (x, x+H]} Σ_{d₁ | r, d₂ | r+2, [d₁,d₂] > z^{u₀−δ}} w(d₁,d₂) = o(H/ln²z)\nfor every x — the definition of the problem with the easy part removed, and\nno instrument for it.\n\n### 4.4 The mean square is not improvable into the arrow\n\nI1 is already square-root cancellation at every level, uniform in s\n(B ≤ 9A²(E−1) with no s in it, `redteam-0829-theorem1.md` §1). Improving B\nto O(1) (measured flat 1.3833–1.4883, `attack-AB-bounded.md` §1.4) would\nchange ms_m by O(ln ln z/ln z) and nothing in the ask column of §2. The\ncorpus's row \"Lemma V's mean-square form as the missing factor — CLOSED\" is\nconfirmed from this side: C1–C2 are a theorem, and they are not where the\nexponent is lost.\n\n### 4.5 The fixed-smooth-profile form is not the arrow, from both ends\n\nWritten in the form the Kloosterman-fraction instruments prove: for smooth\ncompactly supported g₁, g₂ and arbitrary 1-bounded α, β,\nΣ_{d₁, d₂} g₁(d₁/D₁) g₂(d₂/D₂) α(d₁) β(d₂) e(−2h d̄₁/d₂) e(−hx/(d₁d₂)) ≪\n(D₁D₂)^{γ}·H^{o(1)}, uniformly in x, with γ ≤ 0.827142 needed against\nBettin–Chandee's 0.970624 (`rho-maximal-law.md` §5; `sift-limit-attack.md`\n§7b). It is disconnected from REC at both ends. (i) The instruments need a\nsmooth factor in the modulus and the inverted variable; the Rosser weight\nμ(d)·1_{S^±}(d) has none and cannot be decomposed into any: its\nlow-frequency share sits at the random-sign noise floor at 31 of 31 slices\n(`smoothness-front.md` §4.3; `REFUTED.md` \"manufacturing the smooth\nprofile\", CLOSED). (ii) The consumer cannot take a smooth g either:\nBrüdern–Fouvry's pointwise inequality needs λ⁻ ≤ 1_{rough} ≤ λ⁺, which\nMöbius-signed Rosser supports supply and a smooth profile does not. (iii)\nGranting the smoothness anyway, the published bound prices at exponent\n5.0907 > β₂ and holds for x ≪ H^{1.21} against a period of exp(H^{0.2344})\n(`rho-maximal-law.md` §5). So \"at a fixed smooth profile\" the statement is\na different statement with a closed hypothesis, not REC in other clothing.\nRung: CITED throughout; nothing re-priced here.\n\n## 5. Proof gap or truth gap: the arrow against the data\n\n**Against the sharp form first, as the brief asks.** The TPC-implying sharp\nlaw is sup|R_H| ≤ √⟨R_H²⟩·√(2 ln W) at the operative window H ≈ 0.5 z², with\nconstant 1; it fails at z = 19 by 0.6% and on 528 of 640 windows at z = 31\n(`rho-maximal-law.md` §4; `rho-exact-z31-01.md` §4.5). REC is at H = z^{u₀}\nwith u₀ > 2 and allows a factor z^{ask} in place of √(2 ln W): in the table\nof §2 the allowance ask exceeds the Gaussian factor gauss = log_z(2√(2 ln W))\nat every level from z = 19 at u₀ = 4.0 (0.8420 against 0.7905 at z = 19,\n0.9553 against 0.7392 at z = 37) and at u₀ = 4.25, while at u₀ = 3.0 and 3.5\nit does not (0.3408–0.4551 and 0.5918–0.7053 against 0.7392–0.7905). So at\ncomputable z the u₀ ≤ 3.5 forms ask for less than Gaussian recovery and are\nthe forms a sharp-law failure could touch; the u₀ ≥ 4.0 forms sit above the\nGaussian line there, and the sharp law's death at z = 19 says nothing about\nthem. Asymptotically ask → u₀/2 − o(1) while gauss → 0, so every form ends\nabove the Gaussian line; the finite-z ordering is a statement about where\nthe polylogs bite.\n\n**At every computable level the arrow holds, with a certified margin.**\nSince sup|R_H| ≤ 2 sup|ρ̃| (exact), F(z, u₀) = (HM − 1)/(2 sup|ρ̃|) is a lower\nbound on the true margin. From S1 (rows with H < W only):\n\n| u₀ | F at z = 19, 29, 31, 37 | log_z F |\n|---|---|---|\n| 3.00 | 1.478e+1, 2.166e+1, 1.548e+1, 1.360e+1 | 0.9148, 0.9134, 0.7977, 0.7228 |\n| 3.50 | 6.462e+1, 1.168e+2, 8.625e+1, 8.276e+1 | 1.4157, 1.4137, 1.2980, 1.2230 |\n| 4.00 | 2.819e+2, 6.290e+2, 4.803e+2, 5.035e+2 | 1.9160, 1.9137, 1.7980, 1.7230 |\n| 4.25 | 5.885e+2, 1.460e+3, 1.133e+3, 1.242e+3 | 2.1660, 2.1637, 2.0480, 1.9730 |\n\nThe margin in exponent falls with z at every u₀, by 0.19 from z = 19 to 37\nat u₀ = 3.0 and 4.0 alike: in the §2 table the truth column rises from\n0.9873 to 1.2874 while need rises only from 2.9033 to 3.0104 (u₀ = 4.0), and\nu₀ enters need as an additive constant, so the fall is the same at every\nu₀. Rung: MEASURED on the cited exact suprema; the factor 2 in\n|R_H| ≤ 2 sup|ρ̃| is up to 2× conservative and was not tightened (sup|R_H|\nat H = z^{u₀} was not walked).\n\n**Proof gap, not truth gap, at every z where the question can be put.** No\ncomputable level contradicts REC at any u₀ ≥ 3, and the sharp-form failure\nis at a window two exponents below the legal band. Asymptotically: the\ntruth's slope is 2.7660 ± 0.2120 (S5, seven points, estimator returns\n2.500000 on a synthetic z^{2.5} truth), leave-one-out 2.6180 ± 0.2601, and\nthe room under u₀ reads 0.2340 (1.10 se) at u₀ = 3.0, 0.7340 (3.46 se) at\n3.5, 1.2340 (5.82 se) at 4.0, 1.4840 (7.00 se) at 4.25. The F4 envelope\n2√⟨ρ̃²⟩√(2 ln W) itself has slope 2.5111 ± 0.1775 on the same points (S5),\nso the measured truth is tracking the Gaussian envelope, not outrunning it.\nTwo honest limits on that reading: seven points under one octave decide no\nasymptote, and the corpus records local slopes here rising with z\n(`rho-exact-z31-01.md` §0). The u₀ = 3.0 form is therefore the one that\ncould be false; the u₀ = 4.0 form is measured true with 1.2 exponents to\nspare and no known mechanism against it. Neither is a theorem, and the\ndistance from measured truth to proof is the whole of §4.\n\n## 6. What would falsify the arrow, and whether it has run\n\n- **REC(3.0, u₀) false at a computable z**: an exact sup|ρ̃|(z) with\n 2 sup|ρ̃| ≥ HM − 1 at H = z^{u₀}, or, sharper, a walked sup|R_H| at that H\n above HM − 1. RUN at z = 13..37 through the cited exact suprema (S1): no\n failure at any u₀ ≥ 3.0; margins in §5. NOT RUN for sup|R_H| directly at\n H = z^{u₀} (the factor 2 stands in), and NOT RUN at z ≥ 41 (the z = 31, 37\n walks cost 6928 s, `rho-exact-z31-01.md`; z = 41 is a period of 7.4e12\n positions, well above the four-hour rule, so it is an ask).\n- **The truth's slope reaching u₀**: the seven-point slope 2.7660 ± 0.2120\n (S5) against u₀. RUN. Decides nothing asymptotically; u₀ = 3.0 is 1.10 se\n away and u₀ = 4.0 is 5.82 se away. The next exact point moves this more\n than anything else here.\n- **A worst-position construction**: a position x with a large-modulus\n hit count of order H/ln²z (the §4.3 reduced form). The corpus's closest\n test is `sift-limit-attack.md` §7b (3): at a CRT-planted doubly-smooth\n position the absolute sum reaches 0.978 of the trivial bound while the\n signed sum stays about H (max 6 over 20,000 positions at pair count\n 588). RUN at one level; it passed; it is one level.\n- **The floor crossing β₂**: if th(nP(z)) ever exceeded β₂, no law of any\n strength could deliver and REC would be moot rather than false. MEASURED\n ≥ 2.0106 at z = 47 (prefix bound); undecided asymptotically\n (`rho-maximal-law.md` reading 6, two models, neither called).\n- **The cap lemma of 4.1 wrong**: a modulus e with Σ*_a |Θ_e(a)||S_H(a/e)|\n above Vabs(e)·e(ln e + 1), or u_cap below the cited u_sup at some z. RUN\n at z = 13, 17 over every primitive frequency (no violation) and at seven\n levels against the cited column (u_cap ≥ u_sup at all seven, S2).\n- **The reduction of 4.3 wrong**: CAP_{≤ z^{u₀−δ}} not below HM − 1 for\n large z. The inequality is a one-line consequence of 4.1's second bound\n and cannot fail asymptotically; at (4, 0.5) it reads 0.39–0.53 and rising\n over z = 19..37 (S3), so at computable z it is the polylog, not the\n exponent, that is being watched.\n\n## 7. Readings\n\n1. **STATED, one open arrow.** From Lemma V's proven mean square to an\n exponent below β₂ the chain is C1–C6 of §2; C1, C2, C5, C6 are PROVEN,\n C4 is standard and measured (M ln²z = 0.3359–0.3772 at z = 13..37, S0),\n and C3 = REC(s, u₀) is the only open arrow. Every proven substitute for\n C3 lands at exponent 2s + o(1) or pays e^{θ(z)/q}.\n\n2. **ISOLATED.** REC(s, u₀): for fixed s > 1 + √e and fixed u₀ ∈ (2, β₂),\n sup_x |R_H(x)| ≤ z^{u₀/2 − ε}·⟨R_H²⟩^{1/2} at H = z^{u₀} for all large z.\n Its truth gives G₂(z#) ≤ z^{u₀}. It is F2 with the rms factored out, a\n level-of-distribution statement at θ_total = 2s/u₀ measured against the\n remainder's own rms, legal at fixed u₀ > 2 and TPC-strength at u₀ = 2\n or uniformly down to it. It is on no REFUTED row (§3, kill checks).\n\n3. **PROVEN, small, an asymptote for a closed route.** Every absolute-value\n accounting across the lattice is capped by CAP(z) ≤ z^{2s}(2s ln z + 1)·\n 6Π(1 + 2/p)², so u_sup ≤ 2s + o(1), which at s ↓ 1 + √e is 5.2974, the\n trivial-level exponent. Measured u_cap 3.4103 → 4.0593 against the cited\n u_sup 2.0617 → 3.0125 (S2); the cap is loose by 28.62–36.94× against the\n exact Fourier ℓ¹ sum at z = 13, 17. `lemmaV-sup-extension.md`'s\n asymptotic prose at lines 485-486 and 508-516 (2^{π(z)} moduli; a\n C^{π(z)} theorem as the reachable end) cannot hold at fixed s; its\n verdict does (`verify-0830-usup-convention.md` §3).\n\n4. **PROVEN, a reduction.** REC needs proving only for moduli in\n (z^{u₀ − δ}, z^{2s}]; the rest is ℓ¹-controlled below HM − 1 for large z.\n At (4, 0.5) the controlled part reads 0.3948–0.5346 of HM − 1 over\n z = 19..37, rising (S3). The reduced arrow is a bound on the signed count\n of large-modulus lattice hits in every window, with no instrument.\n\n5. **NEGATIVE, each at a named place.** Term-by-term accounting dies at the\n term count (4.1); position-union and hypercontractive routes die at the\n atom √p and the union price e^{θ/q} (4.2); window smoothing and the P(y)\n lever move 0–27% and 1.7–32% of CAP at computable z and no exponent\n (4.3, S3, S4); B is not where the exponent is lost (4.4); the\n smooth-profile form is disconnected from the consumer at both ends and\n priced at 5.0907 even if granted (4.5).\n\n6. **MEASURED: proof gap, not truth gap, at every computable z.** Certified\n margin F = (HM − 1)/(2 sup|ρ̃|) at u₀ = 4: 2.819e+2, 6.290e+2, 4.803e+2,\n 5.035e+2 at z = 19, 29, 31, 37 (log_z F 1.9160 → 1.7230); at u₀ = 3.0:\n 1.478e+1 → 1.360e+1 (log_z F 0.9148 → 0.7228). The truth's slope 2.7660 ±\n 0.2120 sits 1.10 se under u₀ = 3.0 and 5.82 se under u₀ = 4.0 (S5); the\n F4 envelope's slope is 2.5111 ± 0.1775 on the same points. Seven points\n under one octave; asymptotically undecided, with the u₀ = 3.0 form the\n one the data could yet refute.\n\n7. **UNCHANGED.** RML(α) open at every α; no exponent moved; the item's\n named riders (\"Lemma V re-enters as ASK not import; fixed-smooth-profile\n binds\") stand, and §4.5 says why the second is not a property of the\n working point.\n\n## 8. Not reached\n\n- sup|R_H| walked directly at H = z^{u₀} (the factor 2 in |R_H| ≤ 2 sup|ρ̃|\n stands in; at most 2× conservative).\n- Any exact point at z ≥ 41 (period 7.4e12; above the four-hour rule).\n- ⟨ρ̃⁴⟩, the named stopping object of `attack-rhoms-01.md` §3; a\n sub-Gaussian fourth moment (measured m₄/m₂² = 2.32–2.92 there) would cut\n the union price to e^{θ/4}, which still diverges, so it was not pursued.\n- The s-freedom: everything is s = 3.0. The cap's asymptote 2s says a\n smaller s is cheaper for the trivial accounting; whether REC is easier at\n s near 1 + √e was not measured.\n- The u₀-dependence of the reduced arrow's mean square (the sub-lattice\n B_big ≤ 9A²(E−1) holds by the same proof, since dropping terms only lowers\n T(e); its measured size was not tabled).\n- Re-pricing the Kloosterman imports at the ρ-configuration (carried over\n from `rho-maximal-law.md` §5 with its caveat).\n\n*Gate: `node research/qc.js --full` result recorded in the closing report;\nthe producer's tail verifies under `qc/embed.js --check` (code-sha256\nmatches, out-sha256 matches, body bit-honest).*\n", "research/history/staging/attack-0830-comb-tail.md": "# Prime-comb equidistribution in the tail regime: the defect measured exactly at @11..@29, and what the Comb Discrepancy Lemma's own method says there\n\n\n\n> **RIDER 2026-08-30 (orchestrator, from `redteam-0830-engine.md`, which\n> reproduced the tail defect at @11–@23 and the 419,328-term Legendre split\n> with zero exactness failures).** Two corrections: the main-term factor is\n> DERIVED from ∏(1 − 1/q_i) but MEASURED against ∏(1 − 1/(q_i − 1)) — which is\n> also what `natal-cap-28…js`:283 and `certificate-engine.md` §1 compute\n> against — a 0.9% difference at @23 that moves nothing certified; and §30's\n> \"F₁ ∈ [1.5, 4]\" misquotes its source, which says [1.5, 3] and makes no\n> statement at s < 1.\n\n**2026-08-30. TODO item 8, remaining part (b). Staging note, HELD, no\nadversarial pass. Producer: `attack-0830-comb-tail.js`, embedded (121 output\nlines), every anchor asserted before a new number prints. No live file was\ntouched. Every number below is quoted from that OUTPUT block or from an\nembedded artifact cited by file, and every inequality's direction is stated.**\n\n## 0. What is open, and only then what moved\n\n- **Nothing is proven here and no exponent moves.** The deliverable is a\n finite-level table with the tail's defect measured for the first time, a\n derivation of why the Comb Discrepancy Lemma's method cannot be carried\n into the tail, and a located non-closing step. That is the base rate the\n brief named, and it is what happened.\n- **The head is still the only certified region.** The tail's mass is 30.16%\n of Σcap₂ at @23 and 25.56% at @29 (OUTPUT, SUMMARY column 4), falling with\n x; certifying it would lift the certified share from the engine's 22.93%\n to at most 53.09% at @23 [ARITHMETIC on two OUTPUT figures]. No instrument\n in this corpus delivers that certificate at any finite level: the tightest\n legal upper bound in the tail at the certificate's depth is loose by\n F₁ ∈ [1.5, 4] (`attack-0830-buchstab-deep.md` §6 [CITED], re-derived in §5\n below), against a budget of 0.34% of the tail's own mass at @23.\n- **Part (b) splits three ways and two of the three are already settled\n elsewhere.** At K = 0 and shallow K in the limit it is Theorem C\n (`thm-capK-bv.md` §2, PROVEN, empty at every run level). At full depth in\n the tail (q_K → q) the naive statement with main term δ_K is FALSE on the\n record (B_true = 0.9044 at q = 4001, `attack-0830-buchstab-deep.md` SEC 3\n [CITED]) and its corrected form is 8(a), CLOSED as a sieve route on\n 2026-08-30. What is left to (b), and is measured here, is the finite-level\n form at K = 0 and at the certificate's own depth K*. §6.\n- **What moved.** (i) The tail defect E(q) = cap₂(q) − s − dP·(π(A) + π(B)\n − 2π(q−1)) is computed exactly at every tail prime at @11..@29 (9 to 7589\n primes per level), together with its Legendre decomposition over the mids\n into 2^{k+2} prime-in-AP terms, split by modulus range. (ii) The same at\n depth K* against the δ_{K*} main term. (iii) The engine's head-share and\n tail-envelope figures reproduce to the printed digit as custody. §3.\n\n## 1. The statement, as `attack-0830-buchstab-deep.md` §6 scopes it (deliverable a)\n\nLevel x, W = x#, mids 7 ≤ p ≤ x, k = #mids, scour primes x < q ≤ √W. Tail:\nq³ > W + 1, where every admissible cofactor m of a fresh victim v = qm is\nprime with m ≥ q (`thm-mod30-tail.md` §2(a) via `staircase-note.md` Lemma 1,\n[CITED]). With A = ⌊(W−1)/q⌋, B = ⌊(W+1)/q⌋ and T for either,\n\n cap₂(q) − s(q) = #{m prime, q ≤ m ≤ A, qm ≡ 11,17 (30), qm ≢ −2 (p) ∀ mids}\n + #{m prime, q ≤ m ≤ B, qm ≡ 13,19 (30), qm ≢ +2 (p) ∀ mids},\n\nand at depth K the K freshness classes m ≢ ∓2q⁻¹ (mod q_i), i ≤ K, are\nadded (one class per prime, `thm-capK-bv.md` §1 [CITED]). **Statement\n(b), K = 0:** cap₂(q) − s = dP·(π(A) + π(B) − 2π(q−1))·(1 + η(q)) with\ndP = (1/4)∏_{mids}(1 − 1/(p−1)), and Σ_{tail}(cap₂ − s)·|η| small against\nthe floor N − Σ_q cap_{K*}(q), uniformly in the tail. **At depth K:** the\nsame with dP replaced by δ_K = dP·∏_{i≤K}(1 − 1/(q_i − 1)). The quantifier\nthe certificate needs is **for every x** at **K = K*(x)**, and only the\n**upper** half, cap_K(q) ≤ main·(1 + η), enters the pigeonhole floor\nN − Σ cap_K (the legal direction, `attack-wrongdirection-audit.md` §1\n[CITED]; an upper bound on a cap is a lower bound on survivors).\n\n**Rung.** At K = 0 and q_K = W^{o(1)}: PROVEN in the limit, ineffective,\nempty below x = 263 (Theorem C, `thm-capK-bv.md` §4 [CITED]). At K = K*\nwith q_{K*} a power of T: OPEN, HEURISTIC in the engine. At full depth:\nthe δ_K form is REFUTED on the data and the B-corrected form is 8(a), CLOSED.\nFinite level, any K: MEASURED only (this note and the Tail Envelope\nMeasurement).\n\n**What it would buy the engine, from the engine's own ledger.** The certified\nhead carries 9.69 / 17.40 / 22.93% of Σcap₂ at @17 / @19 / @23\n(`natal-cap-28` OUTPUT [CITED], reproduced here to the digit). The tail\ncarries 43.32 / 35.81 / 30.16% of Σcap₂ and 42.65 / 33.61 / 26.70% of\nΣcap_{K*} at the same levels, 25.56% and 21.06% at @29 (OUTPUT SUMMARY). A\nproven tail bound with error η would add the tail's share to the certified\nshare at that level, so at most 22.93 + 30.16 = 53.09% at @23, and the\nbudget it must meet is floor / Σ_{tail}(cap_{K*} − s) = 4841 / 1413605 =\n0.34% at @23 and 31327 / 30129475 = 0.10% at @29 [ARITHMETIC on OUTPUT\nfigures]. The tail's share falls with x (75% at @11 to 25.56% at @29), so\nthis is a finite-level gain that shrinks, not a route.\n\n## 2. Legality and direction\n\nEvery statement attacked below is a finite computation or a per-prime\nasymptotic for a κ = 1 sifted prime count; none is a window statement, so\nAxis A of the wrong-direction test does not apply. The all-x K = 0 form is\nalready a theorem in the limit, hence not TPC-strength. The full-depth\nform is a count of primes m ≤ T with qm ± 2 ∈ P₂, parity-blocked\n(`bv-import-survey.md` §3.3 [CITED]); whether it is TPC-strength is NOT\nDETERMINED here and it is not attacked here. The only inequality used in\nthe certificate's direction is cap_K(q) ≤ main·(1 + η): an upper bound on\nthe cap, which lowers Σcap and raises the floor, legal.\n\n## 3. The finite-level measurement (deliverable b)\n\nCustody first, all asserted in the producer before any new figure prints:\nΣcap₂ = 56 / 880 / 16135 / 308401 / 7034588 at @11..@23 (cap-08, cap-28\nCHECK); tail counts and tail Σcap₂ = (9, 42) / (29, 485) / (105, 6989) /\n(396, 110439) / (1638, 2121291) / (7589, 51660643) at @11..@29 (cap-08\nbands, cap-11, cap-18 OUTPUT); CERT n = 0/1/3/6 with shares 0.00 / 9.69 /\n17.40 / 22.93%, PRIME agg π-form −1.05 / −0.49 / 0.01 / 0.01% and maxrel\n8.33 / 15.26 / 26.31 / 26.59% (cap-28 OUTPUT, matched as printed strings);\nthe nine @11 per-prime tail rows (cap-08); floors 34 / 110 / 82 / 1877 /\n4841 at K* = 0/0/2/10/27. All pass. At @29 the head is not enumerated (the\nlpf array would be 2.1·10⁸ entries) and Σcap_{69} = 143107823 is cited from\ncap-18's floor.\n\n**The tail defect at K = 0** (OUTPUT, SUMMARY; percentages of the tail main\nterm Σ dP·(π(A) + π(B) − 2π(q−1))):\n\n| level | tail n | E signed | Σ\\|E\\| | max \\|E\\|/cap₂ at cap₂ ≥ 20 | sign split +/− | head Σ\\|cap₂ − mainC\\| (all head q) |\n|---|---|---|---|---|---|---|\n| @11 | 9 | −6.03% | 14.92% | 0.00% (no q with cap₂ ≥ 20) | 4/5 | 5.744% |\n| @13 | 29 | 1.09% | 6.50% | 8.33% | 16/13 | 1.252% |\n| @17 | 105 | 0.50% | 2.85% | 15.26% | 54/51 | 0.694% |\n| @19 | 396 | −0.01% | 1.42% | 26.31% | 183/213 | 0.911% |\n| @23 | 1638 | −0.01% | 0.60% | 26.59% | 813/825 | 0.999% |\n| @29 | 7589 | −0.00% | 0.25% | 37.41% | 3772/3817 | not enumerated |\n\n**Reading, calibrated MEASURED.** Per prime the tail defect is LARGER than\nthe head's: its worst relative error grows with level (8.33% to 37.41%,\nattained at small caps of order 20 where a prime count fluctuates by\nits own square root) against 0.01 to 0.40% at the certified head (cap-28\nOUTPUT). In aggregate it is SMALLER from @23 on: Σ|E| is 0.60% of main at\n@23 and 0.25% at @29 against the head's 1.00% at @23, and it falls roughly\nby half per fold while the head's holds near 1%. The signed aggregate is\nbelow 0.01% from @19 on, the sign split is even at every level, and the\nsigned value is inside the floor's budget of §1 at @19, @23 and @29 while\nthe unsigned one is not (0.60% against 0.34% at @23, 0.25% against 0.10%\nat @29). So an aggregate certificate is consistent with the data and a\nper-prime one is not, at every level that exists.\n\n**At depth K*** (OUTPUT, \"tail K=K*\" lines): Σ_{tail}(cap_{K*} − s) against\nthe δ_{K*} main term reads −6.03 / 1.09 / 0.39 / −0.19 / 0.00 / −0.07%\nsigned and 14.92 / 6.50 / 3.33 / 2.42 / 1.08 / 0.54% unsigned at @11..@29,\nand the engine's product form (cap₂ − s)·∏_{i≤K*}(1 − 1/(q_i − 1)) gives\nB_tail = 1.0000 / 1.0000 / 0.9989 / 0.9983 / 1.0001 / 0.9994. At K* the\ntail is in the regime where the Buchstab factor is 1 to four digits,\nalthough Theorem C's hypothesis q_K = T^{o(1)} is formally violated there\n(ln q_{27}/ln T = ln 131/ln T runs 0.38 to 0.51 across the @23 tail\n[ARITHMETIC]).\n\n## 4. The Comb Discrepancy Lemma's method in the tail (deliverable c)\n\n**4.1 The lemma as the engine uses it, carried into the tail, is true and\nsays nothing, and it fails on the main term before the error term.** The\nCertified-Head Theorem's bound |cap₂(q) − mainC(q)| ≤ 2^{j+1}(2·3^k + 1),\nmainC = s + (A + B)·(N/W)·∏_{i 2, and\nan asymptotic only as s → ∞ (the fundamental lemma, whose factor is below 1\nfirst at x = 263 even at K = 0, `thm-capK-bv.md` §4 [CITED]). The\ncertificate's budget is 0.34% at @23 and 0.10% at @29 (§1), against a\nloss factor of 1.5 at best.\n\n**Grade: a proof gap, not a truth gap, on the data.** At K* the δ_{K*} main\nterm matches the tail to 0.00% signed at @23 and −0.07% at @29 with\nB_tail within 0.002 of 1 at every level (§3), so the asymptotic the step\nwould prove is consistent with everything measured; what is missing is an\ninstrument for a κ = 1 asymptotic at bounded s below the sifting limit,\nwhich the corpus has searched for at κ = 2 (`REFUTED.md`, β₂ row) and which\ndoes not exist at κ = 1 either below s = 2 (Iwaniec's linear sieve is the\nsharp instrument, and its f₁(s) = 0 for s ≤ 2 is the statement). It becomes\na truth gap only at full depth, where the δ_K form is refuted by B_true =\n0.9044 (q = 4001, @23 [CITED]); that is 8(a)'s object and is CLOSED. Whether\nthe certificate's depth stays at ρ ≈ 0.42 to 0.56 for all x, or drifts\ntoward the full-depth regime, is the open question of `u2-engine-depth.md`\n(measured u = 4.191 → 3.557 and falling, ANSWERED as not yet at 2 [CITED]).\n\n## 6. Kill check: does (b) reduce to 8(a) or to a REFUTED row\n\nPartly, and the note marks it. (i) The full-depth tail statement IS 8(a)'s\nobject at tail q (the Buchstab factor between the product and the count),\nCLOSED as a sieve route, 2026-08-30; nothing here reopens it. (ii) The\nshallow-K limit statement IS Theorem C, PROVEN and empty. (iii) The\nfinite-level and K* form measured here is neither: it is what\n`certificate-engine.md` §2 calls the Tail Envelope Measurement, now with the\nper-prime defect, its Legendre anatomy and its K* form on the record. No\n`REFUTED.md` row covers it; none is proposed, since a measurement is not a\nroute. Status of 8(b): PARTIAL, and the part that is still open is a proof\ngap at bounded sifting parameter that no instrument in the corpus addresses.\n\n## 7. Brief errors and defects noticed in passing\n\n- \"the tail regime's blocks are longer and fewer\": fewer holds, longer does\n not (§4.2).\n- \"head share is certified at 28-29%\": that is `comb-discrepancy-tight.md`\n §5's per-dilation figure, scratchpad-grade with no embedded OUTPUT; the\n engine's embedded figure is 22.93% at @23, and the sum 53.09% in §1 uses\n the embedded one.\n- `certificate-engine.md` §2 says the K = 0 tail statement \"is exactly the\n statement above\" of Theorem C; at finite level the statement the engine\n uses is at K = K*, where q_{K*} is already T^{0.38..0.51} at @23 and\n Theorem C's hypothesis does not hold. Wording; nothing downstream moves.\n- The engine's tail envelope quotes \"aggregate |err| 0.0-1.1% of tail mass\":\n that is the signed aggregate. The unsigned one is 6.50 / 2.85 / 1.42 /\n 0.60% at @13..@23, and it is the unsigned one a per-prime certificate\n would have to pay.\n\n## 8. What would falsify this, and whether the check has run\n\n| claim | falsifier | check |\n|---|---|---|\n| the per-prime defect E(q) is computed correctly | a mismatch with cap-08's @11 rows, the tail Σcap₂ anchors, or cap-28's PRIME agg and maxrel strings | RUN, all asserted, six levels |\n| the Legendre decomposition is exact | signed term sum ≠ cap₂ − s, or signed defect sum ≠ E(q) | RUN, asserted at every tail prime, tolerance 10⁻⁶ |\n| short terms obey the lemma's < 1 bound | any short term with \\|Δ\\| ≥ 1 | RUN, asserted at every short term |\n| the main-term factor e^γ(1/u − q^{1−u}) | the exact ratio departing from it by more than the finite corrections | PARTIALLY: measured 10-12% above at @23-@29 with the corrections named, not separately computed |\n| B_tail = 1 at K* | a level where Σ_{tail}(cap_{K*} − s) departs from the product by > 1% | RUN at six levels, worst 0.0017 |\n| the non-closing step is a proof gap | a tail prime at K* where the δ_{K*} main term is off by an amount a sieve constant could not absorb | NOT RUN per prime at K*; only the aggregates are on the record |\n| no @29 head figure | none computed | flagged, not a claim |\n", "research/history/staging/attack-0830-rec-cheapest.md": "# REC at its cheapest legal point: the saving asked is z^{1.031}, and the pointwise floor of the certificate says no accounting can supply it\n\n\n\n> **RIDER 2026-09-04 (orchestrator, from `redteam-0904-floor-growth-2.md` §5\n> and `redteam-0904-item0.md`).** §4.2's p* window (4D^{1/27}, D^{1/9}/4) is\n> a convenience bound, neither necessary nor sufficient for family membership\n> at reachable levels: on exact integers the construction's onset is\n> z = 1.875e6 at this note's s = 2.698721 (the analytic 1.055e6 is early by\n> 1.78×) and 4.39e5 at s = 3.0. The floor's crossing of H = z^{u₀} depends on\n> both s and u₀ (5.6e31 is the (2.698721, 4.2165) cell; 10^15.3 at (3.0, β₂)).\n> The growth half survived a second adversarial pass at every admissible\n> level split (1643 splits, minimum margin 0.64); rung unchanged, DERIVED.\n\n> **RIDER 2026-08-30 (orchestrator; both halves red-teamed on independent\n> code).** EXACT half (`redteam-0830-floor-sign.md`): STANDS as PROVEN and is\n> stronger than stated in two places (E3's identity holds at every position,\n> 0 mismatches in 20.5M, not only at doubly non-rough points; sup|R₁| = Ω + M\n> exactly, so sup|ρ̃| ≥ (Ω + M)/2, not (Ω − M)/2); E2 needs the hypothesis\n> D ≥ z this note omits (it fails at s = 0.8; free in the legal band); Ω\n> reproduced at all 26 tabled values and extended to 285, 441, 684, 990,\n> 1155 at z = 79, 83, 89, 97, 101. Three sentences REFUTED as written: E1\n> (L164) \"every pair with gcd | 2 is realised\" — exactly half are, the\n> missing half being the pairs where 2 divides one side only; the ledger's\n> and §0's \"251 against z^{u₀} = 1.05e6\" — 1.05e6 is HM(u₀), z^{u₀} at\n> z = 73 is 7.188e7, so the gap is 68.3× larger than stated; §4.4's reason\n> for λ⁺(P(z)) = 0 (0.24 exceeds 1/8; the closing figure is 0.1169–0.1199,\n> under 1/8 by 4–7%, a near-miss). The exact half constrains REC at NO u₀ by\n> itself, every exactly known log_z Ω being below 2. GROWTH half\n> (`redteam-0830-floor-growth.md`): six attacks, no break, stays DERIVED;\n> hypothesis s ≤ 3 added to the 16s/9 threshold; \"0.012 → 0.35, rising\" is\n> the wrong description (falls 2.5× first); q ≈ z^{5.00} not z^{4.8}; the\n> blind slope test tests the construction, not the law. Ruling: REC, F1, F2\n> and RML below β₂ are false at rung derived-and-red-teamed-once, the route a\n> truth gap at that rung (`REFUTED.md`, 2026-08-30), its falsity beyond\n> z ≈ 5.6e31.\n\n*(2026-08-30 wave, staging. Producer:\n`research/history/staging/attack-0830-rec-cheapest.js`, embedded via\n`qc/embed.js`. No existing file edited, no git command run. Every number\nbelow is in the producer's OUTPUT block or is quoted from an embedded\nartifact cited by file and section. Calibration per claim: PROVEN, MEASURED,\nDERIVED (a written argument not yet adversarially checked), HEURISTIC, OPEN,\nREFUTED. HELD: no adversarial pass has run on §4.)*\n\n## 0. What is open, first\n\n- **Nothing here lowers any exponent, and no route closes.** At the cheapest\n legal point the five routes of `attack-0829n-rml-proof.md` §4 die at the\n same statement they died at yesterday (§2), and the mixed strategy of the\n brief is worth z^{0} (§3). On the brief's kill rule this is PARTIAL.\n- **The new part is adverse to the item.** §4 shows the absolute-value\n accounting is sharp up to a polylog at one window that can be written\n down, so the saving REC asks for, z^{1.031} at the cheapest point and\n z^{2s − u₀} anywhere, is not a saving a proof has failed to find but a\n saving that is not there. The exact half of that (E1–E4 of §4.1) is\n PROVEN and machine-checked at z ≤ 73; the growth half (§4.2, exponent\n 16s/9 − o(1), and 2s − o(1) in the limit) is DERIVED and HELD for one\n adversarial pass. If it stands, REC(s, u₀) is false at every u₀ < 16s/9,\n which at s ↓ 1+√e is 4.7088 > β₂, so no legal point survives; F1, F2 and\n RML(α) at every α < β₂ go with it, and rider 2's floor th(nP) exceeds β₂\n asymptotically. That is the event yesterday's §6 listed under \"the floor\n crossing β₂: REC would be moot rather than false\".\n- **At every computable z it remains a proof gap.** Ω(z), the pointwise\n floor, is exact to z = 73 and reads 251 against z^{u₀} = 1.05e6 (S3); the\n certified lower bound at z = 5e5 is 8.86e17 against 1.07e24 (S4b), a\n factor 1.2e6 short, with the deficit falling 3.01 → 1.65 exponents over\n 601..5e5 and needing to reach 0.58 before any window fails. No computation\n in reach falsifies REC; the falsity is a statement about the limit.\n- **What was not done.** No exact sup at s near 1+√e beyond z = 29 (S2), no\n exact Ω beyond z = 73, no six-prime chains in the certified family (they\n would raise the exponent toward 2s and are dormant below z ≈ 1e5), no\n literature search for the pointwise-size fact, which is standard.\n\n## 1. The cheapest point, written out (deliverable a)\n\nFix η > 0 and set s = 1 + √e + η, u₀ = β₂ − η, D = z^s, H = ⌊z^{u₀}⌋,\nW = P(z). REC(1+√e+η, β₂−η) reads: there are ε > 0 and z₀ with\n\n sup_{x ∈ ℤ/W} |R_H(x)| ≤ z^{(β₂−η)/2 − ε} · ⟨R_H²⟩^{1/2} for all z ≥ z₀,\n\ns and u₀ fixed before z, u₀ never a function of z (the quantifier discipline\nof `attack-0829n-rml-proof.md` §3, unchanged). Its truth gives G₂(z#) ≤\nz^{β₂−η}, an exponent movement of exactly η. The point is legal for every\nη ∈ (0, β₂ − 2): u₀ − 2 = 2.2165 at the working η (S0), fixed and never sent\nto 2 (`attack-wrongdirection-audit.md` §3.10, row 10). It is the cheapest\npoint in one currency and the dearest in the other, and both are read off\nS0 (constants 1+√e = 2.648721270700, β₂ = 4.26645028414864, difference of\nthe doubles 2(1+√e) − β₂ = 1.030992):\n\n| η | s | u₀ | cap-currency ask 2s − u₀ | rms-currency ask u₀/2 | bought |\n|---|---|---|---|---|---|\n| 0 (limit, illegal: M → 0) | 2.6487 | 4.2665 | 1.0310 | 2.1332 | 0.00 |\n| 0.01 | 2.6587 | 4.2565 | 1.0610 | 2.1282 | 0.01 |\n| 0.05 (working point) | 2.6987 | 4.2165 | 1.1810 | 2.1082 | 0.05 |\n| 0.10 | 2.7487 | 4.1665 | 1.3310 | 2.0832 | 0.10 |\n| 0.35 (≈ yesterday's s = 3) | 2.9987 | 3.9165 | 2.0810 | 1.9582 | 0.35 |\n\n- **Cap currency.** Against any absolute-value accounting (C3-triv, C3-cap of\n yesterday's §2, both PROVEN at exponent 2s + o(1)) the saving asked is\n z^{2s − u₀ − o(1)} = z^{1.030992 + 3η}. At η → 0 this is the\n 2(1+√e) − β₂ gap of `sift-limit-attack.md` §4.5 and §7b (K_FH =\n 5.297442541400 against β₂), the brief's z^{1.031}: correct to the digits\n the brief gives. It is the smallest saving any legal point asks for, and it\n buys the least, η.\n- **Rms currency.** Against the PROVEN mean square (C1–C2) the recovery\n asked, sup/rms ≤ z^{u₀/2 − ε}, is LARGEST here: u₀/2 = 2.1332 − η/2 against\n 1.5 at u₀ = 3. Cheap in one accounting is dear in the other, and the\n routes of §2 live in different currencies, which is why they have to be\n priced separately.\n- **Main term at the point.** M ln²z is MEASURED 0.3332 → 0.2715 at z = 13..47\n at s = 2.698721 (S2, S3; the s = 3 family reads 0.3359–0.3772, S1), falling:\n the linear-sieve main term ln(s−1)² − 1 vanishes at s = 1+√e, so the\n consumer's HM − 1 is itself thinner here than anywhere else in the band.\n\n## 2. The five routes at a demand of z^{1.031} (deliverable b)\n\nEach route is re-priced at (s, u₀) = (2.698721, 4.216450) and at η → 0.\n\n1. **Term-count cap (yesterday §4.1).** The cap delivers exponent 2s + o(1)\n by taking absolute values across the lattice, so its saving over itself\n is identically 1 at every (s, u₀). A demand of z^{1.031} is smaller than\n the z^{2.0} demand at (3, 4), but the route supplies zero saving at any\n positive demand, so it does not close and cannot close by lowering the\n demand. Measured at the point: u_cap is not re-tabled; §4 shows the cap\n is sharp up to a polylog, which is stronger than \"not improved\".\n2. **Position union and moments (yesterday §4.2).** Price e^{θ(z)/q} at\n moment order q. Against the rms-currency ask z^{u₀/2}: with Gaussian-shaped\n moments ‖R‖_q ≤ C√q · rms the route needs e^{θ(z)/q}·√q ≤ z^{u₀/2 − ε},\n i.e. q ≥ θ(z)/((u₀/2) ln z)(1 + o(1)). The cheapest point asks the largest\n u₀/2, so the moment order needed is the SMALLEST of the band, z/(2.1332\n ln z) against z/(1.5 ln z) at u₀ = 3: a constant factor 1.42 on an order\n that still diverges. The only proven moment is q = 2 (I1); the failing\n step is a moment bound of unbounded order, the same step as yesterday.\n Against the cap-currency figure z^{1.031} the comparison is not even the\n right one: e^{θ(z)/q} ≤ z^{1.031} needs q ≥ z/(1.031 ln z) and then\n ‖R‖_q is unproven for every q > 2.\n3. **Window smoothing and the P(y) lever (yesterday §4.3).** Polylog movers;\n at the cheapest point the reduced range of moduli (z^{u₀−δ}, z^{2s}] is\n only 1.181 + δ exponents wide, so there is less to remove, and the\n removed part was never the exponent. §4.3 below shows the counterexample\n window lives entirely inside the reduced range.\n4. **The mean square (yesterday §4.4).** I1 is square-root cancellation at\n every level, s-uniform; not where the exponent is lost. Unchanged.\n5. **The smooth-profile form (yesterday §4.5).** Disconnected at both ends;\n nothing at the point changes the hypothesis. Unchanged.\n\n**The single failing step, at the cheapest point:** a bound on the signed\ncount of large-modulus lattice hits in one window, uniform in the window's\nposition, by o(H/ln²z), with H = z^{β₂−η} and the moduli up to z^{2s}. That\nis the statement of yesterday's §4.3, the reduced arrow, at the point where\nthe modulus range is narrowest. It is the same statement, so by the brief's\nkill rule the routes verdict is PARTIAL. §4 then says the statement is false.\n\n## 3. The mixed strategy, with the arithmetic (deliverable c)\n\nSplit the moduli at z^{u₀−δ}. Below it, the ℓ¹ accounting is PROVEN to sit\nunder HM − 1 for large z (yesterday reading 4). Above it, the only two\ninstruments are the cap (sup ≤ CAP_big ≤ z^{2s} polylog, PROVEN) and the mean\nsquare (⟨R²_big⟩ ≤ B_big H, PROVEN, almost-all only). A sup bound from the\npair {sup ≤ A, rms ≤ B} is exactly A: with no moment above the second, the\nmean square contributes nothing to a bound at one position (Axis C of\n`attack-wrongdirection-audit.md` §1: the density of bad positions it\ncontrols is W·B²/λ², never zero). So every mixture \"part from the mean\nsquare, part from a cap over a restricted range\" reduces to the cap on the\nbig moduli:\n\n saving = CAP_big / z^{2s} = z^{o(1)}, for every split point δ,\n\nbecause the ℓ¹ mass sits at the top of the modulus range (yesterday S3:\nthe share of CAP above z^{3.5} at s = 3 reads 0.5684 at z = 37 and rising).\nThe mixed strategy reaches z^{0}, not z^{1.031}. There is no arithmetic to\ndo beyond this, and §4 makes the zero structural: the cap's exponent is the\ntrue exponent of the certificate at a planted window.\n\n## 4. The pointwise floor of the certificate (deliverable e, the new part)\n\nEverything in §§2–3 dies at the same statement as yesterday. What is new is\nthe reason: the absolute-value accounting is not loose. Its exponent, up to a\npolylog, is the true size of the certificate at one window, and the window\ncan be written down.\n\n### 4.1 Four exact facts\n\nWrite λ^±(n) = Σ_{d | n, d ∈ D^±} μ(d) for the Rosser weights of level D\n(`sift-limit-lemmaV.js` `rosserSupport`, the pilot's own support), P = P(z),\nand for a position r let n₁ = gcd(r, P), n₂ = gcd(r+2, P). Then:\n\n- **(E1) cc(r) is a function of (n₁, n₂) alone**, cc(r) = λ⁻(n₁)λ⁺(n₂) +\n λ⁺(n₁)λ⁻(n₂) − λ⁺(n₁)λ⁺(n₂), and every pair (n₁, n₂) of divisors of P with\n gcd(n₁, n₂) | 2 is realised by some r in ℤ/W (CRT: r ≡ 0 mod n₁,\n r ≡ −2 mod n₂, r ≢ 0, −2 mod every other odd prime). PROVEN, one line.\n- **(E2) λ⁺(n) ≥ 0 and λ⁻(n) ≤ 0 at every n | P with n > 1.** The Buchstab\n boundary identity: for a prefix-closed support D, Σ_{d | n, d ∈ D} μ(d) =\n 1_{n = 1} − Σ μ(d₀)·1[(n, P(p_m)) = 1], the sum over first-exit chains\n d₀ = p₁ ⋯ p_m | n (p₁ > ⋯ > p_m, p₁ ⋯ p_{m−1} ∈ D, d₀ ∉ D). D⁺ exits only at\n odd m (its conditions are p₁ ⋯ p_{m−1} p_m³ ≤ D at odd m; the size cap is\n implied at even m), so every boundary term has μ(d₀) = −1 and λ⁺(n) is a\n count; D⁻ exits only at even m, so λ⁻(n) is minus a count. Checked\n exhaustively over every divisor of P(z) at z = 13..73 (S3, `viol` = 0 at\n every level, both s). PROVEN.\n- **(E3) At a doubly non-rough point, cc(r) = −(A₁A₂ + A₁B₂ + B₁A₂) with\n A_i = λ⁺(n_i) ≥ 0, B_i = −λ⁻(n_i) ≥ 0**, so cc(r) ≤ −A₁A₂. From (E1),\n (E2). PROVEN. (Where r+2 is rough, cc(r) = λ⁻(n₁) = −B₁; the S3\n maximisers at z ≤ 19 are of that kind, B₂ printed as −1.)\n- **(E4) For every window (x, x+H] containing r: T(x) ≤ cc(r) + (H − 1)**,\n because cc(r′) ≤ θ(r′)θ(r′+2) ≤ 1 pointwise (Brüdern–Fouvry, PROVEN,\n `rho-maximal-law.md` §2 (iv)). PROVEN.\n\nDefine **Ω(z, s) = −min_r cc(r) = max over splits of (A₁A₂ + A₁B₂ + B₁A₂)**.\nThen, for every H ≤ Ω(z, s): the window starting at r − 1 has T ≤ −1, so\n**nP(z) > Ω(z, s)** (rider 2's floor), **F1 fails at H**, and since\nR_1(x) = cc(x+1) − M, **sup|ρ̃| ≥ (Ω − M)/2**. Direction guard: every one of\nthese is a lower bound on a remainder or an upper bound on the certificate,\nthe adverse direction; none says anything about G₂ itself and none has TPC\ncontent.\n\n### 4.2 The exit-chain count, and its size\n\nBy the boundary identity, for n = P₁ with smallest prime p*,\n\n λ⁺(P₁) = #{ d′ ∈ D⁺ : d′ | P₁/p*, ω(d′) even, d′ > D/p*³ },\n\nthe chains that exit D⁺ exactly when p* is appended (the odd-m condition\nd′·p*³ ≤ D fails). Take the sub-family ω(d′) = 4, d′ = p₁p₂p₃p₄ with\np₁³ ≤ D and p₁p₂p₃³ ≤ D (the two D⁺ conditions, both prefix conditions\nimplied). At the extremal exponent pattern (s/3, s/3, s/9, s/9) the product\nreaches D^{8/9}, and it exits at any p* > (D/d′)^{1/3} ≈ D^{1/27}. Dyadic\nranges p₁ ∈ (D^{1/3}/2, D^{1/3}], p₂ ∈ (D^{1/3}/4, D^{1/3}/2], p₃ ∈\n(D^{1/9}/2, D^{1/9}], p₄ ∈ (D^{1/9}/4, D^{1/9}/2] satisfy both conditions\nidentically, give d′ ≥ D^{8/9}/64, and exit at every p* > 4D^{1/27}. Split\nthe primes in each range by index parity into two halves and give each half\nits own p*: two disjoint sets P₁, P₂, each with A_i ≍ D^{8/9}/ln⁴D by\nChebyshev's bounds on the four dyadic prime counts, whence\n\n Ω(z, s) ≥ A₁A₂ ≫ z^{16s/9} / ln⁸ z (DERIVED, HELD).\n\nThe same construction with 2k primes at the pattern (s/3, s/3, s/9, s/9, …,\ns/3^k, s/3^k) gives exponent 2s(1 − 3^{−k}), so **Ω(z, s) ≥ z^{2s − o(1)}**\nfor fixed s as z → ∞, the o(1) being O(3^{−k}) at the cost of a polylog\nln^{−4k} z. The 4-chain exponent already suffices for what follows.\n\nWhat this is not: it is not new mathematics. That combinatorial sieve weights\ntake polynomially large values at smooth integers, and that absolute\nremainder accounting is therefore sharp for sifting short intervals, is the\nstandard reason a level of distribution is needed at all. The corpus measured\nthe absolute side of this on 2026-08-18 (`sift-limit-attack.md` §7b (3), the\nplanted doubly-smooth n₀ at 0.978 of the trivial bound) and read the signed\nside as small because at z = 13 it is: Ω(13) = 1 (S3). The signed side grows,\nand the reason is (E2): at a planted point every term of cc has the same\nsign, because both weights are counts of exit chains.\n\n### 4.3 What it does to REC, F1, F2, RML and the floor\n\n- **REC(s, u₀) is false for every u₀ < 16s/9, if §4.2 stands.** REC implies\n min_x T(x) ≥ 1 at H = z^{u₀} for all z ≥ z₀ (chain C3–C6, PROVEN), and\n (E4) with Ω ≥ z^{16s/9 − o(1)} gives a window with T ≤ −1 for all large z\n once u₀ < 16s/9. At s = 1+√e+η the threshold is 16s/9 = 4.7088 + 1.78η\n and the cheapest point u₀ = β₂ − η sits 0.4424 + 2.78η below it (S0);\n at s = 3 the threshold is 5.3333. So the whole legal band (2, β₂] is\n below the threshold at every admissible s, and the \"cheapest\" point is\n not cheaper in the only currency that turns out to matter.\n- **F1 and F2 (`rho-maximal-law.md` §1) fail at every H < Ω(z)**, including\n the per-position form ∀x ∃H: the window starting at r − 1 has T ≤ −1 for\n every H ≤ Ω. **RML(α) fails for every α < 16s/9** through sup|ρ̃| ≥\n (Ω − M)/2, which is above β₂: the named legal target of item 0 cannot be\n true at any sufficient α. **Rider 2's floor th(nP) ≥ log_z Ω → ≥ 16s/9 >\n β₂**: the deliverable band [th(nP), β₂) is asymptotically empty. This\n decides the \"two models, neither called\" of `rho-maximal-law.md` reading 6\n in the direction of the room closing, not by either model but by a\n construction.\n- **Where the counterexample lives.** The pairs (d₁, d₂) = (d′p*₁, d″p*₂)\n have q = d₁d₂ ≈ D^{16/9}·p*₁p*₂ ≈ z^{4.8}, inside yesterday's reduced\n range (z^{u₀−δ}, z^{2s}] and above the window: the reduced arrow of §4.3\n there is exactly where the certificate is negative.\n- **What is NOT claimed.** Nothing about G₂(z#) itself moves: a certificate\n failing at one window says nothing about the count of admissible r in\n that window (the measured G₂ exponent stays ~1.5). Nothing about the mean\n square moves (planted windows are a vanishing fraction of the W\n positions; I1 stands). The Gaussian ρ law measured TRUE at z ≤ 37\n (`rho-maximal-law.md` §4) is a finite-z fact in the regime where the\n chains are dormant (Ω(37, 3.0) = 21 against 2 sup|ρ̃| − M = 104.410, S3).\n\n### 4.4 The numbers behind §4.2\n\n**Exact, both s (S3).** Ω(z, 2.698721) = 1, 2, 3, 6, 10, 18, 22, 30, 45, 63,\n86, 111, 134, 168, 205, 251 at z = 13..73, log_z Ω from 0 to 1.2878; the\nsplit enumeration reproduces the walked min cc at z = 13..29 (five MATCH\nlines), λ⁺(P(z)) = 0 at every level (the D/8 slab is empty: max d/D ≤ 0.24\neverywhere, so the two-prime and four-prime chains cannot reach it at these\nz), and the maximisers are one band of middle primes with p* = 5 or 7\nagainst the top three to five primes. At s = 3.0, Ω = 1, 2, 3, 3, 3, 9, 21,\n36, 63, 100 at z = 13..47 and Ω − M ≤ 2 sup|ρ̃| holds at all seven cited\nlevels (`ok` at 13..37), the control that ties the enumeration to the\ncorpus's exact suprema. Against the consumer, Ω/HM reads 6.4e−5 to 3.8e−4\nat s = 3 and 2.1e−4 to 4.0e−4 at the cheapest s: three and a half orders\nshort of a failing window.\n\n**Hill-climb lower bounds (S4a).** Ω ≥ 6.840e2 at z = 89 rising to 2.108e6\nat z = 601, log_z from 1.4543 to 2.2757, against z^{u₀}: ratio 3.7e−6 to 6.8e−6, no trend.\nThe maximisers switch from band splits to interleaved splits at z ≥ 151\n(A₁, B₁, A₂ all in the hundreds by z = 601).\n\n**Certified sub-family (S4b).** With the primes above 47 split by index\nparity and the best p* ≤ 47 per side, exact counting of 2- and 4-chains\ngives Ω ≥ 9.000e4 at z = 601 (control: the family count 300 equals the\nexact λ⁺ of the same set, so no chain is double-counted) and 8.863e17 at\nz = 5e5, log_z 1.7828 → 3.1493. Four-prime chains take over from z = 2e4\n(1.56e5 of 2.29e5 at z = 2e4; all of it at 5e5, where the two-prime family\nis empty because p* is capped at 47). The deficit 16s/9 − log_z Ω falls\n3.0149 → 1.6485 across 601..5e5, and the bound tracks z^{16s/9}/ln⁸z at\n0.012 of it at z = 601 and 0.35 at z = 5e5, rising: consistent with the\nChebyshev-count derivation and its ln⁸ z, and not decisive for it.\n\n**Fits (S5).** ln Ω on ln z: 3.2825 ± 0.1205 on the exact points z = 29..73,\n4.1394 ± 0.0898 on the hill-climb points 89..601 (4.1033 ± 0.1272 on\n251..601), estimator returning 2.500000 on a synthetic z^{2.5}. The slopes\nrise with z, as the corpus records for every sup-type instrument here\n(`rho-exact-z31-01.md` §0), and the certified family is what the slope is\nrising toward. At s = 3.0 the exact slope over 29..47 is 6.7327 ± 0.5818 on\nsix points, above the cited sup|ρ̃| slope 2.766 ± 0.212 on 13..37: the\npointwise floor is growing faster than the sup it sits under, as it must if\nit is to become the sup.\n\n## 5. The arrow where it can be seen (deliverable d)\n\n**At the cheapest s, exact (S2).** s = 2.698721, H = ⌊z^{4.216450}⌋, full\nperiod, R_H walked directly (no factor 2):\n\n| z | H > W? | sup\\|R_H\\| | rms(R_H) | sup/rms | log_z(sup/rms) | allowance u₀/2 | log_z F |\n|---|---|---|---|---|---|---|---|\n| 13 | yes | 3.312 | 1.0567 | 3.134 | 0.4453 | 2.1082 | 2.5865 |\n| 17 | yes | 5.996 | 1.9121 | 3.136 | 0.4034 | 2.1082 | 2.4801 |\n| 19 | no | 9.649 | 2.4586 | 3.925 | 0.4644 | 2.1082 | 2.3114 |\n| 23 | no | 14.334 | 3.1119 | 4.606 | 0.4871 | 2.1082 | 2.2563 |\n| 29 | no | 27.971 | 5.8294 | 4.798 | 0.4657 | 2.1082 | 2.1679 |\n\nThe recovery the arrow asks for is z^{2.1082}; the measured recovery is\nz^{0.40–0.49}, and min_x T runs 2517 → 41384 against HM 2520 → 41411: every\nwindow certifies, by a margin F = (HM−1)/sup of z^{2.59} falling to z^{2.17}.\nThe z = 13, 17 rows have H > W and are calibration only.\n\n**At s = 3.0, cited (S1).** need − ms = 0.9200 → 1.0765 over z = 13..47\n(the rms-currency ask at these z, smaller than the asymptotic 2.1082 because\nM and the polylogs bite), truth log_z(2 sup|ρ̃|) = 0.6458 → 1.2874 over\nz = 13..37, log_z F = 2.4457 → 1.9394. No exact sup exists at z = 41, 43,\n47 (only ⟨ρ̃²⟩; `rho-exact-z31-01.md` §0 prices those walks at 37× to\n65,232× the z = 37 cost), so the brief's \"z = 13..47\" is 13..37 for the\nsup and 13..47 for the rms. Margins fall with z at both s, by about 0.42 of\nexponent from 13 to 29 at the cheapest s and 0.51 from 13 to 37 at s = 3,\nthe same drift as yesterday's table.\n\n## 6. Proof gap or truth gap, and the falsifiers\n\n**Proof gap at every computable z; truth gap in the limit, on §4.2.** The\ntwo are not in tension: the exact Ω at z ≤ 73 is three and a half orders\nunder HM, the certified bound at z = 5e5 is six orders under z^{u₀}, and the\nstatement REC makes is about all z ≥ z₀, where the construction wins by a\npower of z once the deficit 16s/9 − log_z Ω (1.65 at z = 5e5, falling)\ndrops below 16s/9 − u₀ = 0.5812. What would settle it, and whether it ran:\n\n- **(E2) wrong: a divisor n of P(z) with λ⁺(n) < 0 or λ⁻(n) > 0.** RUN\n exhaustively over all 2^{π(z)} divisors at z = 13..73 (cheapest s) and\n 13..47 (s = 3): zero violations. A failure at larger z would break §4.1\n and the standard theory of the Rosser sieve with it.\n- **(E4) wrong: a window containing r with T > cc(r) + H − 1.** Cannot\n happen while Brüdern–Fouvry's pointwise inequality holds; checked\n implicitly by the S2 walks (min T ≥ Ω-consistent at five levels).\n- **The split enumeration wrong.** RUN: it reproduces the walked min cc at\n z = 13..29 and respects Ω − M ≤ 2 sup|ρ̃| at the seven cited exact levels;\n the certified family at z = 601 equals the exact λ⁺ of its own set.\n- **§4.2 wrong: the 4-chain count in the top slab not ≫ D^{8/9}/ln⁴D.** NOT\n proven here beyond the written argument; the certified counts rise from\n 0.012 to 0.35 of z^{16s/9}/ln⁸z over 601..5e5 (S4b), consistent and not\n decisive. An adversarial re-derivation of the dyadic-range construction\n (the p* > 4D^{1/27} exit, the two prefix conditions, the parity split) is\n the pass this note is HELD for. The claim needs only Chebyshev-type\n bounds on prime counts in dyadic ranges.\n- **A computable failing window.** Would need Ω(z) > z^{u₀}: out of reach at\n any z this producer or the box can hold (the deficit must fall by another\n 1.07 exponents from z = 5e5). NOT RUN and not runnable; this is why the\n verdict is asymptotic.\n- **Against the item's own instruments.** th(nP) measured 1.5963 → 1.8569 at\n z = 13..31 and ≥ 2.0106 at 47 (`rho-maximal-law.md` §2 rider 2) is\n consistent: nP > Ω ≥ 100 at z = 47, s = 3 (S3), far below the measured\n floor z^{2.0106} = 2.3e3, so the floor is not yet set by the chains.\n\n## 7. Readings\n\n1. **STATED (deliverable a).** REC(1+√e+η, β₂−η): cap-currency ask\n z^{1.030992 + 3η}, rms-currency ask z^{2.1332 − η/2}, exponent bought η;\n M ln²z measured 0.3332 → 0.2715 at z = 13..47 there (S0, S2, S3). The\n brief's z^{1.031} verifies.\n2. **PARTIAL (deliverable b).** None of the five routes closes at the\n smaller demand: the cap supplies zero saving at any demand, the union\n route needs a moment of order z/(2.1332 ln z), a factor 1.42 smaller than\n at u₀ = 3 and still divergent, the rest are unchanged. The failing step is\n yesterday's reduced arrow, verbatim.\n3. **ZERO (deliverable c).** A mixed mean-square-plus-cap strategy delivers\n the cap on the large moduli, z^{2s − o(1)}, at every split point: with no\n moment above the second the mean square adds nothing at one position.\n4. **PROVEN, small, and machine-checked.** cc(r) at a doubly non-rough\n point is minus a sum of three products of exit-chain counts, every\n window containing r has T ≤ cc(r) + H − 1, and so nP > Ω, F1 fails below\n Ω, and sup|ρ̃| ≥ (Ω − M)/2; Ω exact at z ≤ 73 (251 at 73, cheapest s) and\n z ≤ 47 (100 at 47, s = 3), controls all pass (S3).\n5. **DERIVED, HELD (deliverable e).** Ω(z, s) ≫ z^{16s/9}/ln⁸z by a dyadic\n four-prime construction, and z^{2s − o(1)} in the limit; then REC is\n false for every u₀ < 16s/9 (4.7088 at s ↓ 1+√e), the cheapest point and\n the whole band with it, RML(α) is false at every α < β₂, and rider 2's\n band is asymptotically empty. Certified lower bounds reach 8.863e17 at\n z = 5e5 (log_z 3.1493), 0.35 of the model and rising (S4b).\n6. **MEASURED: proof gap at every computable z.** Ω/HM ≤ 4.0e−4 at z ≤ 73,\n Ω/z^{u₀} ≤ 8.3e−7 certified at z = 5e5; sup/rms at the cheapest s reads\n z^{0.40–0.49} against an allowance z^{2.1082} at z = 13..29 (S2). The\n slopes of ln Ω rise with z: 3.28 ± 0.12 exact (29..73), 4.14 ± 0.09 on the\n hill-climb range (S5).\n7. **UNCHANGED.** No exponent moved; RML(α) stays OPEN as a question until\n the adversarial pass on §4.2 runs, at which point the item's named target\n is REFUTED rather than open.\n\n## 8. Not reached, and corrections to the record\n\nNot reached: an exact Ω past z = 73 (3^{π(z)} splits; z = 79 is 3× the\nz = 73 cost and z = 89 27×); six-prime chains in the certified family (they\ncarry the 2s − o(1) limit and are empty below z ≈ 1e5 at this s); the\nexact sup|R_H| at s near 1+√e beyond z = 29; a literature citation for the\npointwise size of Rosser weights at smooth integers (standard, not searched\nunder `SEARCH-CONVENTIONS.md`); the s-dependence of Ω beyond the two values\nrun; any re-pricing of the u₀ = 3 and u₀ = 4 forms, which §4.3 covers\nwithout new numbers.\n\nCorrections to the record (proposed, no live file edited):\n\n- Brief, deliverable (d): \"z = 13..47 where the data exist\" — the exact\n sup|ρ̃| exists at z = 13..37 only; 41..47 carry ⟨ρ̃²⟩ alone\n (`rho-exact-z31-01.md` §0, §4.6).\n- Brief, deliverable (b): \"the position-union route, whose price e^{θ(z)/q}\n may or may not beat z^{1.031}\" compares a rms-currency price with a\n cap-currency ask; the union route's ask is z^{u₀/2}, largest at the\n cheapest point (§2 item 2).\n- `attack-0829n-rml-proof.md` §5 and reading 6, \"proof gap, not truth gap\":\n correct at every computable z, and to be read with §4 here for the limit.\n- `rho-maximal-law.md` reading 6, \"two models, neither called\": §4.3 calls\n the room closing, conditional on §4.2 surviving its pass.\n- `REFUTED.md`: a candidate row, HELD until the pass: \"REC(s, u₀), F1, F2 and\n RML(α) below 16s/9 on the Rosser vector-sieve lattice — REFUTED (planted\n exit-chain window: cc(r) = −(A₁A₂ + A₁B₂ + B₁A₂), T ≤ cc(r) + H − 1,\n Ω ≥ z^{16s/9 − o(1)})\".\n\n*Gate: `node research/qc.js --full` result recorded in the closing report;\nthe producer's tail verifies under `qc/embed.js --check`.*\n", "research/history/staging/attack-c2drift-01.md": "# The c2' drift is a diagonal-frame property, not a G2 anomaly — and its shape splits two survivors\n\n\n\n*(2026-08-21. Producer `research/attack-c2drift-01.js`, embedded (1.5 s,\n`--check` passes, 11 self-tests). Every number below is in its OUTPUT block.\nInstrument-first per `exponent-control.md`: every p-value is a rank in a\n4000-replicate null ensemble generated in the same pass on the same grid from\na truth whose answer is known, never a textbook standard error.)*\n\n## 0. The question\n\nThe A144311 adoption left a new open question: c2' = G2/(m·lnD) runs\n[0.4463, 0.5004] on the custody band x = 11..31 and [0.4842, 0.5337] on the\ntrusted tail x >= 41. Why does it drift upward? Three candidates: (i) a\nconstant plus a slowly-vanishing finite-size correction, (ii) genuine growth\nlike a power of lnD or of lnlnD, (iii) the maxgap-law §7a ledger's own\nnext-order term r(x)·ln₃x/(ln₂x)², scored zero-parameter. And the cheapest\ndecisive discriminator, run first: does the one-class column c1 = h/(m1·lnD1)\nover A048670's 64 published terms drift the same way?\n\n## 1. The h-side discriminator: YES — the drift is a property of the diagonal frame\n\nc1 has the same signature as c2', point for point in shape: a high head\n(0.4712 at x = 11), a dip (minimum 0.3359 at x = 59), then a slow rise to\n0.3756 at x = 311. Post-dip the rise is decisive and survives both noise\nmodels:\n\n| window (c1) | n | b per ln ln x | p (lognormal null) | p (Gumbel/lnD null) |\n|---|---|---|---|---|\n| x >= 11 | 60 | −0.1091 ± 0.0335 | 0.9968 | 0.9527 |\n| x >= 17 | 58 | +0.0238 ± 0.0256 | 0.1790 | 0.3643 |\n| x >= 59 | 48 | **+0.1553 ± 0.0261** | **0.0000** | **0.0000** |\n| x >= 101 | 39 | +0.1524 ± 0.0182 | 0.0000 | 0.0000 |\n| x >= 179 | 24 | +0.2187 ± 0.0442 | 0.0000 | 0.0000 |\n\n(The negative full-range slope is the x = 11..13 head, maxgap-law §6's known\noutlier structure.) This independently confirms maxgap-law's (log p)^0.12 with\na different regressor and the full 64 terms, and it settles the question's\nshape: **whatever drives the drift acts on one class and two classes alike.\nEvery twin-specific explanation of \"why does c2' drift\" is dead on arrival**;\nthe object under study is the diagonal law's constant in both sieve\ndimensions, and the c2' version is just the shorter, noisier copy (post-head\nrates: c2' 0.23–0.43 per ln ln x against c1's 0.16–0.22 — same direction,\nsame order, wide errors on the short ladder).\n\n## 2. The instrument, calibrated before the data was read\n\n- **Deterministic recovery** (rounding-only, real 18-term grid): a flat truth\n reads b_PLNX = −0.0140 (no spurious drift from discretisation); a true\n (ln x)^1.0 reads 0.9903; the ledger shape reads b_LEDG = 0.9956. The\n estimator sees what is there.\n- **The bias runs DOWN here, not up.** A pure Gumbel finite-size truth\n (c ∝ 1 + γ/lnD) reads b_PLNX = −0.1341: a Gumbel head term biases the drift\n estimator negative, so the measured drifts are conservative. (The corpus's\n one-class lesson — positive log inside the truth biases a power fit up —\n applies to exponent fits on g itself; on the ratio column the danger\n inverts, and it is now priced.)\n- **Cross-talk, load-bearing:** ln lnD and ln ln x are nearly collinear\n on-range (lnD ≈ θ(x) ≈ x). A true (lnD)^0.12 reads as (ln x)^0.50, a true\n (ln x)^0.30 reads as (lnD)^0.07. **Candidate (ii)'s two subforms cannot be\n separated by shape on any reachable range; only the per-frame rate is\n measurable.**\n- **Power:** against a true (ln x)^0.30 drift at the 5% level, the c2' grid\n (n = 18) has 92.8% power, the c1 grid (n = 60) has 100.0%. Nothing weaker\n than s ≈ 0.3 was detectable on the c2' side — the h-side is where the power\n lives, which is why §1 leads.\n\n## 3. The calibrated verdict on the three candidates\n\n**c2's own drift is real:** x >= 17 slope 0.3351 ± 0.1098 with p_logn =\n0.0053 and p_EV = 0.0075 (significant under both noise models); full-range\nn = 18 is noise-model-dependent (0.0092 vs 0.1737 — the EV null's slope sd is\n0.2056 at that n, honestly reported, not resolved). Deleting the x = 37\noutlier leaves b = 0.2219 ± 0.0578: the outlier does not carry the drift.\n\n| candidate | verdict | the numbers |\n|---|---|---|\n| (i) EV/Gumbel finite-size, c∞·(1 + γ/lnD) | **SIGN-REFUTED** | predicts b_FS = +0.577 (falling column); every fitted b_FS is negative (−0.597 c2' full, −3.52 c1 long lever) |\n| (i)' empirical mirror: negative deficit, c → c∞ from below | **ALIVE, narrowly ahead on the longest lever** | on c1 x >= 59, FS beats both powers (ΔAICc 1.1 vs 6.3–7.2); implied limits c1∞ = 0.3799, c2'∞ = 0.5345, i.e. G2 → 1.28·x·ln²x |\n| (ii) genuine polylog growth | **ALIVE, wins the c2' table** | COMP (γ-term fixed + (ln x)^s) is AICc-best on c2' x >= 17 at s = 0.4335; post-head windows s = 0.23–0.43 ≡ (lnD)^{0.05–0.08}; subforms unresolvable (§2) |\n| (iii) ledger shape, zero-parameter | **REFUTED globally, wins one deep window** | anchored on the custody band it overshoots the trusted tail 5.5× (rms(ln) 0.5874 vs 0.1066 flat; predicts +13.2% over 41→79 against +3.1% observed); fitted shape exponent 0.0894 (c2') and −0.0574 (c1) against the required 1 — yet on c1 x >= 59 the same k = 1 anchored shape is the outright AICc winner (−392.1) |\n\nThe ledger reading deserves its sentence: it fails exactly where its own\nderivation warns (the shape vanishes at x = e^e = 15.15 and is steepest just\nabove), and becomes the best zero-shape-parameter description precisely in\nthe deep window — maxgap-law §7a's \"the flatness is a property of where the\ndata sits on ln₃x/(ln₂x)²\" survives in exactly the regime it claims.\n\nBetween (i)' and (ii) this data cannot decide (1.1 AICc units on the best\nlever). The cheapest future discriminator is on the ONE-CLASS side, not ours:\nunder (i)' c1 saturates at its fitted 0.3799 (c1(311) = 0.3756 is already\n99% of the way); under (ii) it climbs through it.\n\n## 4. The 1d compatibility algebra, exact\n\nG2/x² = c2'·(m·lnD/x²) identically. Measured over the ladder 11 → 79:\n\n Δln(G2/x²) = Δln c2' + Δln(m·lnD/x²)\n −0.2365 = +0.0540 + (−0.2905)\n\nThe drift consumes 18.6% of the frame factor's fall. Since d(ln lnD)/d(ln x)\n≈ 1 on the diagonal, the PLND slope IS the per-ln-x growth rate of c2', and\nG2/x² keeps falling exactly while b_PLND < −dln(m·lnD/x²)/dln x — a threshold\nmeasured at 0.147 on-range and rising to 1 − 2·lnln x/ln x → 1 asymptotically\n(m·lnD → 2.40·x·ln²x). Measured b_PLND = 0.052: a third of the on-range\nthreshold, far below the asymptotic one. Consequences:\n\n- **Every fixed polylog drift keeps G2/x² → 0** — c2' ~ (ln x)^s gives\n G2 ~ x·ln^{2+s}x for any s, the full Maier-Pomerance s = 1 included.\n- **The two trusted drifts are in no tension anywhere in candidate space.**\n They jointly pin the correction between m·lnD and x² to polylog size,\n ln^{2+s}x with s between 0.23 and 0.43 on this range; only a near-unit\n POWER of lnD ≈ x could reverse 1d's lean, and the measured lnD-exponent is\n 0.05–0.08, two orders of structure away.\n- Under candidate (i)' the correction CONVERGES: G2 → 0.5345·m·lnD ≈\n 1.28·x·ln²x, a constant diagonal law with a vanishing deficit — the\n strongest possible \"G2/x² falls\" reading. Under (ii) the polylog compounds\n without limit and still never reaches x². 1d's central question is\n narrowed, not settled: the correction is polylog either way; whether it\n saturates is exactly the (i)'-vs-(ii) split.\n\n## 5. What was NOT reached\n\n- **(i)' vs (ii) is undecided** — 1.1 AICc units on the best lever; the\n designed test (does c1 pass 0.3799?) needs A048670 terms past p = 311,\n which no one has computed.\n- **(ii)'s subforms (lnD-power vs lnlnD-power) are unresolvable on-range** by\n the collinearity of §2 — this is an instrument ceiling, not a data\n shortfall; no reachable ladder length fixes it.\n- **Full-range c2' significance stays noise-model-dependent** (p_EV = 0.17 at\n n = 18); the x >= 17 window verdict (significant under both) is the honest\n quotable.\n- **The ledger's r-driven growth was never testable:** r(x) = 2 everywhere\n reachable (checked at every grid point to 311), so only its T1 shape was\n scored; the r-ladder's predicted breaks live at x ~ 1e6 and beyond.\n- **Mechanism.** Nothing here says WHY the frame drifts — only that the why\n must be class-count-blind, that it is not the Gumbel mean, and that its\n size sits at a tenth of Maier-Pomerance's requirement on both columns.\n- 8 of the 18 c2' terms and the last 6 c1 terms are single-witness\n (Alekseyev/Wang; Bozek); in-house verification remains the paper-phase\n deliverable it already was.\n\n---\n\n*This is a history/staging record: a process document, superseded-in-place by\nlater work. The live claims travel with the producer's READINGS block.*\n", "research/history/staging/attack-hsub-01.md": "# TODO 1d: (H-sub) attacked — three proven reductions, the fold verdict, and no counterexample\n\n\n\n*2026-08-20. Producer: `research/attack-hsub-01.js` (0.1 s,\n`code-sha256 f14f5789`, `out-sha256 ee4e9643`, formal embed with input\nhashing, seven numbered readings; fingerprint verified with `embed.js\n--check` after writing). One defect was found in the producer's own first\nrun and repaired before this report's numbers were read: the proven-floor\ncheck's slack accumulator initialized at 0, capping its printed minimum —\nthe could-not-fail class. The re-embed used `--force` (the tool's\ncorrected-code path); the override is stamped in the tail. Nothing was\ncommitted or pushed; no existing corpus document was edited. HELD, not integrated — awaiting the standing one-pass\nadversarial review. Legend as in `research/sift-limit-attack.md`: **[PROVEN]**\npublished theorem or a proof given here; **[VERIFIED]** checked\ncomputationally here; **[MEASURED]** empirical, finite range; **[INFERRED]**\ndeduction from sourced facts.*\n\n---\n\n## 0. The verdict, up front\n\n> **(H-sub) is not proven and not refuted, but it is now a smaller statement\n> aimed at a wider window, and the pass found one target the trap cannot\n> touch. Three reductions of the lemma's hypothesis are proven; the fold\n> machinery's inability to reach any of them is made exact; and the hunt\n> over all 111 reachable pairs found no drifting family and no\n> counterexample.**\n>\n> 1. **The lemma consumes only power pairs.** Its own eight-line proof uses\n> (H-sub) at the pairs `(b^k, b)` and nowhere else, so the hypothesis\n> weakens to (H-sub-pow) with the conclusion unchanged. The record\n> defects 1.0761 at (4,10) and 1.0330 at (4,12) are not power pairs and\n> stop constraining the route: the trap window widens from\n> `[1.0761, 1.3946)` (0.3185 nats) to **`[1.0033, 1.3946)` (0.3913 nats,\n> trusted)** and `[0.9694, 1.3555)` (0.3861, custody-only). **[PROVEN /\n> VERIFIED]**\n> 2. **One chain carries the TPC face; no chain carries existence.** A single\n> base `b₀` with chain defect `K < S(b₀)` forces `limsup < 2`; limit\n> existence needs chains at inf-realizing bases. And the chains that\n> decide the trap (bases 16 and 66) have **zero reachable pairs** — the\n> decisive family is dataless. **[PROVEN / VERIFIED]**\n> 3. **The doubling slice is trap-free and prize-bearing.** `Ĝ(2s) ≤ C₂·Ĝ(s)`\n> gives `limsup ≤ log₂C₂`; its TPC line `C₂ < 4` is already excluded by\n> custody data (`C₂ ≥ 348/66 = 5.2727`), so no provable doubling constant\n> is TPC-implying through the slice's own conclusion — and every\n> `C₂ ∈ [5.2727, 19.2455)` lowers the proven\n> exponent ceiling 4.2665 to `log₂C₂ ∈ [2.3985, 4.2665)`. A 1.29-nat\n> landing zone with no trap in it. **[PROVEN the implication; VERIFIED\n> the window]**\n> 4. **The fold machinery cannot reach (H-sub), exactly.** The fold step is\n> `×p_next/p ≤ 5/3 < 2` in Ĝ's argument coordinate — below the integer\n> floor `t ≥ 2` — so every admissible pair is multi-fold; (H-sub) is\n> identically a head-dominance statement about the nonnegative\n> fold-budget sequence, and the proven tools bound the defect from the\n> side it does not need. **[PROVEN / VERIFIED]**\n> 5. **No counterexample.** Max `D = 1.0761` at (4,10), min `−0.2400` at\n> (7,7); the (4,t), (s,2) and diagonal families all read bounded against\n> the in-pass constant-defect null. **[MEASURED]**\n\n---\n\n## 1. Reduction 1: the hypothesis shrinks to power pairs, and the floor drops\n\nRe-reading the proof in `attack-fekete-1d-02-lemma.js`: the limsup half\niterates `f(b^{k+1}) = f(b^k·b) ≤ f(b^k) + f(b) + K` — the pair `(b^k, b)` —\nthen bridges intervals with the proven (H-mono); the liminf half uses only\nthe definition of the inf. No other pair is ever consumed. So\n\n> **(H-sub-pow)** `f(b^{k+1}) ≤ f(b^k) + f(b) + K` for all integers `b ≥ 2`,\n> `k ≥ 1`\n\nimplies the full conclusion — `lim f(n)/ln n = inf_n (f(n)+K)/ln n` exists,\nand `β < 2 ⟺ S(n) > K at one integer` — by the existing proof, character for\ncharacter. **[PROVEN, by inspection of the existing proof]**\n\nThe consequence is a data statement. Only 15 power pairs are reachable\n(9 custody), and the family's sup defect is\n\n| grade | floor | at | rests on |\n|---|---|---|---|\n| trusted | **1.0033** | (16,4) — base-4 chain, k = 2 | `Ĝ(64) = G₂(61#) = 1080` (Wang 2024 a(18) — the same term as the 1.3946 ceiling) |\n| custody | **0.9694** | (16,2) — base-2 chain, k = 4 | `G₂(31#) = 348`, `G₂(13#) = 66`, corpus-exact |\n\n(4,10) and (4,12) are not of the form `(b^k, b)` (verified), so the two\nworst defects on the 111-pair table no longer bound the constant the route\nneeds. **The trap window is now `ln C ∈ [1.0033, 1.3946)` at trusted grade —\n0.3913 nats, 23% wider than the standing 0.3185 — and `[0.9694, 1.3555)`,\n0.3861 nats, on custody terms alone.** Ceilings are unchanged: they belong\nto the conclusion formula (`max_b S(b)`), not to the pair family.\n**[VERIFIED]**\n\n## 2. What a cofinal chain buys — the exact answer to the composition question\n\nThe task asked whether a proof on a cofinal sub-family, composed along the\nladder, feeds the lemma. Split by face:\n\n- **Limsup face (which is the TPC face).** (H-sub-pow) at ONE base `b₀`\n gives `limsup f(n)/ln n ≤ (f(b₀)+K)/ln b₀`; so `K < S(b₀)` on one\n geometric chain forces `limsup < 2`. More generally a cofinal chain with\n two-sidedly pinned steps `a·n_j ≤ n_{j+1} ≤ b·n_j` (`a > 1`) and\n `f(n_{j+1}) ≤ f(n_j) + f(b) + K` gives `limsup ≤ (f(b)+K)/ln a`. Steps\n pinned only from above lose the denominator and give nothing. **[PROVEN]**\n- **Existence face.** No single chain, and no base subset `B`, gives limit\n existence: `limsup ≤ inf_{b∈B}(f(b)+K)/ln b` but `liminf ≥ L` takes the\n inf over ALL `n`, and the two close only when `B` realizes the inf —\n unknowable in advance, so the clean sufficient family is all-bases power\n pairs (Reduction 1) and nothing smaller. **[PROVEN]**\n- **The decisive chains are dataless.** No base `b ≥ 10` has a single\n reachable pair (`b² > 82`), so the chains at the threshold bases 16 and 66\n — the ones whose `K < S(b₀)` would be TPC — are constrained by no\n measurement the corpus owns; the trap-window floor comes entirely from\n bases 2..9. **[VERIFIED]**\n- **The \"fold-step family\" (s a primorial level, t the next prime) does not\n exist inside the candidate.** See §4(i): the fold step is a sub-integer\n multiplier, and the corpus's own domain correction (real bases die at\n `ln C ≥ 2.01`, `import-interp.md` §4) forbids restating the candidate\n down there. **[INFERRED, from verified ingredients]**\n\n## 3. The slice ladder: per-slice traps, and the doubling window\n\nFixing `t₀` and quantifying over `s` (the slice contains its own base chain\n`(t₀^k, t₀)`) gives `limsup ≤ (f(t₀)+K_{t₀})/ln t₀`: TPC-implying iff\n`K_{t₀} < S(t₀)`; improves the proven `β₂ = 4.26645` (`dhr-verification.md`\nrow 1a) iff `K_{t₀} < β₂ ln t₀ − f(t₀)`. Measured floors against both lines\n(full table in the producer):\n\n| t₀ | reachable sup D | S(t₀) | trap | improvement ceiling |\n|---|---|---|---|---|\n| 2 | 0.9694 (custody) | 0.6931 | **closed by data** | 2.2641 |\n| 3 | 0.5810 | 0.4055 | closed | 2.8954 |\n| 4 | 1.0761 | 0.9808 | closed | 4.1228 |\n| 5 | 0.7697 | 0.7340 | closed | 4.3817 |\n| 6 | 1.0415 (custody) | 1.0986 | **OPEN, width 0.0572** | 5.1595 |\n| 7..13 | — | — | all OPEN, wider | — |\n\nTwo headlines fall out:\n\n**The doubling window (t₀ = 2) — the pass's sharpest find.** In constant\nform the slice is `Ĝ(2s) ≤ C₂·Ĝ(s)` and the conclusion is\n`limsup ln G₂(x#)/ln x ≤ log₂C₂`. Custody data already force\n`C₂ ≥ Ĝ(32)/Ĝ(16) = 348/66 = 5.2727 > 4`, so **no provable doubling constant\ncan be TPC-implying — the explicit-constant trap that guards every other\nface of this route is void on the doubling slice**, and a soft proof is not\nexcluded by trap logic. Simultaneously every `C₂ ∈ [5.2727, 2^{β₂} =\n19.2455)` lowers the proven exponent ceiling: `limsup ≤ log₂C₂ ∈ [2.3985,\n4.2665)`. One inequality — a per-doubling growth cap on the twin-gap ladder\n— with a 1.29-nat-wide landing zone in which every point is consistent with\nall data, non-TPC, and an unconditional improvement of the corpus's\nstrongest proven upper bound. **[PROVEN the implication; VERIFIED the\nwindow; the inequality itself is OPEN]**\n\n**The t₀ = 6 slice trap is nearly shut.** Any explicit\n`K₆ ∈ [1.0415, 1.0986)` — 0.0572 nats, floor at the custody-exact (4,6) —\nis TPC-implying, with implied bound 1.9681 at the floor. This does not\ndisplace the general trap window; it records that a single fixed-t\ninequality at t = 6 is already TPC-strength on a margin five times thinner\nthan the general one. **[VERIFIED]**\n\n## 4. Structure first: the exact reason the fold machinery falls short\n\n**(i) The fold step is not an integer pair.** In Ĝ's argument coordinate the\nfold `G₂(p_n#) → G₂(p_{n+1}#)` is multiplication by `p_{n+1}/p_n ≤ 5/3 < 2`\n(max at 3→5; verified at all 21 steps). The candidate's floor `t ≥ 2` —\nforced when real bases died at `ln C ≥ 2.01` — sits strictly above the\none-fold regime, so every admissible pair crosses\n`π(st) − π(s) ≥ π(2s) − π(s) ≥ 1` folds (Bertrand), unboundedly many as `s`\ngrows (the (4,10) pair already spans 10 folds; verified minimum 1, at\n(2,2)). A constant per-fold cap compounds to `exp(#folds)`: no one-fold\ntheorem composes into (H-sub). **[VERIFIED on the range; the Bertrand floor\nis PROVEN]**\n\n**(ii) (H-sub) in fold coordinates is head dominance.** With\n`r_1 = ln G₂(2#)` and `r_i = ln[G₂(p_i#)/G₂(p_{i−1}#)] ≥ 0` — nonnegative\nby the PROVEN (H-mono) — the identity\n`f(st) − f(s) = Σ_{π(s) < i ≤ π(st)} r_i` holds (verified to max dev\n1.78e−15 at all 111 pairs), so (H-sub) is exactly: **the budget spent in the\nwindow `(s, st]` never exceeds the budget of the first `π(t)` folds\nplus K.** The window holds ~`st/ln st` folds against the head's `π(t)`; the\ncandidate is a statement that late fold budgets thin out, which is global\nregularity, not local structure. The three worst pairs decompose cleanly\n(e.g. (4,10): 10-fold window budget 4.4773 vs 4-fold head 3.4012, excess =\nD = 1.0761). **[PROVEN the identity; VERIFIED]**\n\n**(iii) The proven tools point the wrong way.** The copy theorem and\n(H-mono) (folding deletes slots, deletion merges gaps) and the block-ladder\nrecord's disjoint-set super-additivity all give LOWER bounds on `Ĝ(st)` —\nthey prove `D(s,t) ≥ −min(f(s), f(t))` (one line from (H-mono); holds at\nevery pair with strictly positive slack, minimum 0.6931 at (2,3), and the\nmeasured min D −0.2400 sits far above its floor −3.4012 at (7,7)). The\nMirror-Sweep Lemma is symmetry structure within ONE fold\n(alignment `a` and `w−a` degrade equally); it halves fingerprints and\nlocates seam anomalies but carries no magnitude bound and nothing that\ncomposes across folds. No named sub-family of (H-sub) is provable with the\ncorpus's current machinery, and this pass proves none. **[PROVEN the floor;\nINFERRED the assessment]**\n\n## 5. The hunt: no counterexample, no drift\n\nOver all 111 reachable integer pairs (`st ≤ 82`, trusted 22-term ladder,\ncross-checked against `attack-fekete-1d-01`'s headline numbers to four\nplaces):\n\n- **Extremes.** Max `D = 1.0761` at (4,10); the next four are (4,6) 1.0415,\n (4,12) 1.0330, (6,12) 1.0202, (4,16) 1.0033. Min `D = −0.2400` at (7,7);\n 7 of 111 pairs are negative (the super-multiplicative side). **[VERIFIED]**\n- **The (4,t) family** — carrier of both record defects — peaks at t = 10\n (the 37-spike) and every `t ≥ 13` sits below the peak. Slope against\n `ln t`: `+0.0808 ± 0.0957` raw, `−0.0020 ± 0.1143` against the in-pass\n constant-defect null (POW, same primes, same step construction).\n **Consistent with bounded; no drift.** **[MEASURED]**\n- **The doubling family (s,2)**, s = 2..41: `−0.0336 ± 0.0596` against the\n null. Consistent with bounded. **[MEASURED]**\n- **The diagonal (b,b)**, b = 2..9 — the one place an unbounded defect could\n live under a power-log law: reads `−0.5188 ± 0.3966` BELOW the\n constant-defect null (1.3 σ, 8 points), i.e. leaning toward the\n decreasing-defect (super-multiplicative) side. The diagonal remains blind\n past `√82 ≈ 9`, exactly as `attack-fekete-1d-01` reading 7 recorded; this\n pass adds data consistent with \"not growing\" and cannot exclude growth.\n **[MEASURED]**\n\nAn honest \"the defect grows along family F\" would have killed the route; no\nreachable family shows it.\n\n## 6. Trap grading of everything this pass produced\n\nNo explicit K was derived (nothing here proves an instance of (H-sub)), so\nnothing trips the trap. What moved are the WINDOWS, graded per the standing\nrule against `[1.0761, 1.3946)` trusted / `[1.0761, 1.3555)` custody:\n\n| object | window | grade | status vs trap |\n|---|---|---|---|\n| power-pair candidate (Reduction 1) | `ln C ∈ [1.0033, 1.3946)` | floor rests on Wang a(18); ceiling likewise | trap WIDENED 0.3185 → 0.3913 nats |\n| power-pair candidate, custody | `ln C ∈ [0.9694, 1.3555)` | fully custody | 0.3861 nats |\n| single-chain TPC face (base 16 / 66) | any `K < 1.3555 / 1.3946` | custody / trusted | TPC-implying; family dataless |\n| t = 6 slice | `K₆ ∈ [1.0415, 1.0986)` | custody | TPC-implying, width 0.0572 |\n| doubling slice | `C₂ ∈ [5.2727, 19.2455)` | custody floor | **trap-free**: TPC needs `C₂ < 4`, excluded by data; every point improves 4.2665 |\n\n## 7. NOT REACHED\n\n- **No instance of (H-sub) proven, for any family.** The three reductions\n are quantifier surgery on the lemma — they change what must be proven, not\n what is proven. The route's gap is smaller and better lit, still open.\n- **The doubling inequality `Ĝ(2s) ≤ C₂·Ĝ(s)` is posed, not attacked.** It\n is the natural next target (trap-free, prize-bearing); this pass did not\n attempt a proof and does not know whether one is soft.\n- **No new defect data.** The reachable pair set is capped at `st ≤ 82`\n until the ladder grows; the first riser enumeration (61#) remains\n ~146,969× the 43# run.\n- **The diagonal stays invisible past b = 9.** Growth of the true defect on\n the diagonal is neither seen nor excluded; the -01 blind spot stands.\n- **de Bruijn–Erdős Theorem 22 remains unopened**, consistent with the\n bounded reading; nothing here cites its content.\n- **Whether the reductions compose** (e.g. power pairs at prime bases only,\n giving the prime-base threshold `max_q S(q) = 1.2946` trusted) was\n computed in exploration but not carried into the producer; the all-bases\n form of Reduction 1 is the operative statement.\n\n## 8. Sources and reproduction\n\n| artifact | what it carries |\n|---|---|\n| `research/attack-hsub-01.js` | everything above: the three reductions (proofs in the header), the power-pair table, the slice ladder, the fold-budget identity, the traces with in-pass nulls (§§1–6) |\n| `research/attack-fekete-1d-02-lemma.js` | the lemma whose proof Reduction 1 inspects; (H-mono) proven + verified |\n| `research/attack-fekete-1d-01-defect47.js` | the 111-pair table this pass cross-checks; the threshold freeze |\n| `research/history/staging/fekete-1d.md`, `redteam-0820-math.md` §1 | the route's state and the trusted-ladder trap window this pass widens |\n| `research/history/staging/import-interp.md` §§3–4 | the defect identity; the real-base death that forces `t ≥ 2` |\n| `research/history/staging/attack-0c-holesweep.md` §4 | the Mirror-Sweep Lemma assessed in §4(iii) |\n| `research/dhr-verification.md` row 1a | `β₂ = 4.26645028414864191641`, the proven ceiling the doubling window lowers |\n| `research/exact-g2-ladder.js`, `research/import-interp-01-bgt-defect.js` | the two ladders, parsed at run time, input-hashed by the embed |\n\nReproduce with `node research/qc/embed.js --check research/attack-hsub-01.js`;\nthe fingerprint matches as of 2026-08-20.\n", "research/history/staging/c2prime-refit-22.md": "# The 22-term refit, and why c₂′ drifts: the control drifts the same way, and ten points cannot measure the rate\n\n\n\n*(2026-08-28. Producer `research/history/staging/c2prime-refit-22.js`, embedded\n(0.1 s, `--check` passes, 12 custody guards). Every number below is in that\nfile's OUTPUT block. Scratchpad-grade: no ensemble nulls are run here, and the\nspreads quoted are the control column's own sliding-window distribution, not\np-values. TODO item 1c, first move.)*\n\n## 0. First: the brief is partly wrong, and the corrections come before the work\n\nThree premises of item 1c do not survive contact with the repo.\n\n**The 22-term refit is not entirely undone.** `exponent-control.md` §5 was\nalready refit on the trusted A144311 ladder on 2026-08-21 (script S11): raw\nx-frame 1.777 ± 0.029, θ-frame 1.647 ± 0.021, control-corrected central 1.498,\nbracket 1.3 to 1.8. Those are the file's headline numbers and they are 22-term\nnumbers today. What was still at ten terms is narrower and is what this file\nrefits: the §2 model-discrimination table for G₂ (quoted again in §7 item 1),\nthe §4 structural column Qg, and the §6 G₂ margin trend.\n\n**c₂′ is not a coefficient in a fitted exponent model.** It is a per-x ratio\ncolumn, c₂′ = G₂(x#)/(m·lnD) with D = ∏_{2\n\n**TODO item 8(c), \"tighten 2·3^k (~30× slack)\". 2026-08-28.\nScript: `research/history/staging/comb-discrepancy-tight.js`, 27.7 s, all asserts\npassed. SCRATCHPAD-GRADE: the script carries no embedded OUTPUT block, so every\nnumber below is quoted as a measurement of a finite deterministic quantity, with\nthe command that produced it, and not as a custody artifact.**\n\n```\nnode --max-old-space-size=4096 research/history/staging/comb-discrepancy-tight.js\n```\n\n## 0. What is open, and what failed\n\n**The engine channel is where the item lives, and it is the channel that moved\nleast.** The certificate engine (`natal-cap-28-analytic-certificate.js` RESULT 1)\nprices *dilated* combs, so it needs a constant uniform over the dilation family.\nThe construction proven below, maximised over that family, is **worse** than the\nstanding seeded transfer of `level-ledger-tight.md` §6c: 91.79 against 26.68 at\n@17, a factor 3.44 in the wrong direction. The grouping gain is a property of the\nα = 1 classes, not of the family. Recorded as a negative.\n\n**No exponent moves.** Every bound here still costs a factor near 3 per added\nmid in the worst case; the measured per-fold step of the blocked bound at fixed\nY is about 2.4, and of the sharp constant about 1.82. Nothing here bears on\n`REFUTED.md`'s discrepancy-theory rows, which are closed as routes; this is a\nfinite lemma's constant and stays that.\n\n**The \"~30×\" in the item's title is now an exact number and it is level-dependent,\nnot a constant.** The slack 2·3^k / D_x reads 2.68 / 4.29 / 6.61 / 12.36 / 16.41 /\n28.99 / 53.64 at @7..@29. The 30 that item 8(c) quotes is the @23 value and it is\ngrowing, so \"the slack\" is not a single factor to be taken back once.\n\n**Not evaluated.** The sharp per-dilation constant at @19 and @23 (the exact range\nper dilation) is not computed: at @23 that is 14.9 M candidate steps per dilation\nagainst about 1000 dilations. The blocked bound is used instead, so the engine\nnumbers in §5 are floors.\n\n## 1. The lemma as proven, and where the 2 and the 3^k come from\n\nProducer: `research/natal-cap-25-excess-law.js` header, lines 25-29, restated in\n`certificate-engine.md`'s status table and cited in `level-ledger-tight.md` §6c.\n\nThe natal comb at level x is\n\n C_x = { r : r ≡ 11 or 17 (mod 30); r ≢ 0 and r ≢ p−2 (mod p) for every mid 7 ≤ p ≤ x },\n\nwith W = x#, N = 2·∏(p−2), density ρ = N/W, and k = #mids = π(x) − 3.\n\n**Comb Discrepancy Lemma (PROVEN).** For every integer s and every integer ℓ,\n|#(C_x ∩ [s, s+ℓ)) − ρℓ| ≤ 2·3^k.\n\n*Proof, reconstructed.* Expand ∏_p (1 − [r≡0] − [r≡p−2]) into 3^k signed terms,\none per pair (d, c) with d | M = ∏mids squarefree and c a choice from {0, p−2}\nfor each p | d, with sign (−1)^{ω(d)}. Each term, intersected with one of the two\nunit classes mod 30, is a single arithmetic progression of modulus 30d by CRT,\nso the comb's indicator is a signed sum of 2·3^k progression indicators. An AP\ncount in an interval is off its density share by less than 1. The shares sum to\nρℓ exactly, since Σ_{d,c} (−1)^{ω(d)}·2/(30d) = (2/30)·∏(1 − 2/p) = ρ. ∎\n\n**So the 2 is the mod-30 unit pair and the 3 is the three terms 1, −[≡0],\n−[≡p−2] per mid.** The 2 is *not* the two excluded classes: those are the second\nand third term inside each factor of 3. The exponent is k = π(x) − 3, not\nπ(x) − 1, because 2, 3 and 5 are already spent on the mod-30 pair, which is why\n`level-ledger-tight.md` §6c found this bound the tighter of the two to start with.\n\n## 2. The constant has an exact optimum, and it is computable\n\nWrite G(u) = #(C_x ∩ [0,u)) − ρu on integers u ∈ [0, W]. G is W-periodic, since\nG(0) = G(W) = 0.\n\n**Theorem A (Sharp Window Constant, PROVEN).** For all integers s, ℓ,\n#(C_x ∩ [s, s+ℓ)) − ρℓ = G(s+ℓ) − G(s), hence\n\n sup over integer s, ℓ of |#(C_x ∩ [s, s+ℓ)) − ρℓ| = max G − min G =: D_x,\n\nand the supremum is attained with ℓ ≤ W: take s at the argmin and ℓ reaching the\nargmax, or the reverse for the negative sign.\n\nTheorem A is one line and it says the lemma's constant is not open-ended. D_x is\nthe floor under every bound of this shape, so the only questions left are its\nvalue and whether a formula for it exists.\n\n**Theorem B (Blocked Legendre Bound, PROVEN).** Group the expansion of §1 by the\ndivisor d. For fixed d all 2^{ω(d)} class terms carry the same sign (−1)^{ω(d)},\nand their supports are distinct residues mod d, hence disjoint. They therefore\nmerge into a single periodic set A_d of period 30d carrying exactly 2^{ω(d)+1}\npoints, and\n\n D_x ≤ Σ_{d | M} range(A_d),\n\nwith range taken over integer u. More generally, for any subset Y of the mids\nkept unexpanded,\n\n D_x ≤ Σ_{d | M∖Y} range(A_{d,Y}),\n\nwhere A_{d,Y} has period 30·W_Y·d and N_Y·2^{ω(d)} points, N_Y = 2·∏_{p∈Y}(p−2).\n\nTwo consequences worth stating. First, **bounding each block's range by its own\npoint count 2^{ω(d)+1} returns 2·3^k exactly**, since Σ_{d|M} 2^{ω(d)+1} = 2·3^k.\nThe standing lemma *is* the trivial per-block bound, and the whole of the\navailable gain is the difference between a block's point count and its actual\nrange. Second, the bound costs N_Y·3^{k−|Y|} point-operations, not O(W), so it is\nevaluable at levels no tile enumeration reaches, and it is exact when Y is the\nwhole mid set.\n\nTheorem B works entirely in the original coordinate and never rescales an\ninterval, which is why the integer-u range is the right quantity. That\ndistinguishes it from `level-ledger-tight.md` Theorem 2, whose sub-intervals have\nrational endpoints and which therefore needs the continuous-u range; the two\ndiffer by up to one block density (1.6 against 1.5333 at y = 5).\n\n## 3. Avenue (c), the maximal empty run, is a lower bound and it is weak\n\nA run of length g with no comb member gives a window with count 0 against share\nρg. The quantity that run measures is the maximal *empty* run, one less than the\nmaximal member-to-member gap g_max, since a window of length g_max contains one\nmember and not none: at @7 the comb is {11, 17, 41, 71, 101, 107, 137, 167, 191,\n197}, the maximal gap is 30 and the maximal empty run is 29. So the valid bound\nis **D_x ≥ ρ·(g_max − 1)**, exactly. Measured: ρ·(g_max − 1) = 1.381 / 2.299 /\n2.934 / 4.509 / 5.596 / 7.676 at @7..@23 against D_x = 2.238 / 4.195 / 8.165 /\n13.102 / 29.619 / 50.295; at @29 the corrected value is one ρ (0.0221) below the\n9.425 the ρ·g_max form gives and it is not re-derived here. The maximal empty run\nexplains 61.7% of D_x at @7 and about 11.5% at @29, falling at every level. As a\nroute to the constant it is dominated; as a certificate that D_x cannot be small\nit is cheap and it holds. The overstatement is one ρ and the slack is large, so\nthe printed inequality held numerically at every level and no downstream number\nmoves; the derivation and the percentages do.\n\n## 4. The table\n\nMeasured, `comb-discrepancy-tight.js` Parts 1, 2, 4. \"cap-25 maxE\" is the\nembedded grid maximum from `natal-cap-25-excess-law.js`'s OUTPUT block, which is\na maximum over a scour-length grid and so is a *floor* under D_x; the script\nasserts that ordering at all six levels where it exists.\n\n| level | k | 2·3^k | ET(x) | blocked Y≤5 | blocked Y≤13 | blocked Y≤19 | **D_x (sharp)** | cap-25 maxE | 2·3^k / D_x |\n|---|---|---|---|---|---|---|---|---|---|\n| @7 | 1 | 6 | 9.919 | 3.3 | - | - | **2.2381** | 1.2 | 2.68 |\n| @11 | 2 | 18 | 26.590 | 7.4 | - | - | **4.1948** | 2.7 | 4.29 |\n| @13 | 3 | 54 | 64.948 | 18.3 | 8.2 | - | **8.1648** | 5.9 | 6.61 |\n| @17 | 4 | 162 | 162.044 | 41.0 | 17.1 | - | **13.1021** | 10.1 | 12.36 |\n| @19 | 5 | 486 | 398.659 | 110.3 | 49.0 | 29.6 | **29.6189** | 21.5 | 16.41 |\n| @23 | 6 | 1,458 | not run | 247.7 | 110.1 | 61.2 | **50.2952** | 34.4 | 28.99 |\n| @29 | 7 | 4,374 | not run | 578.9 | 249.4 | 140.9 | **81.5492** | - | 53.64 |\n| @31 | 8 | 13,122 | not run | 1,333.0 | 555.6 | 328.8 | out of reach | - | - |\n| @37 | 9 | 39,366 | not run | 3,246.5 | 1,384.6 | 832.5 | out of reach | - | - |\n\n**Sharpest at every level where it exists: D_x**, by Theorem A, and there is\nnothing below it. Where D_x is out of reach the sharpest is the blocked bound at\nthe deepest affordable Y: 328.8 at @31 and 832.5 at @37 with Y ≤ 19, gains of\n39.9 and 47.3 over 2·3^k. The largest level with an exact answer is **@29:\nD_29 = 81.5492 against 2·3^7 = 4374, a factor 53.64, with the blocked Y ≤ 19\nbound at 140.9, itself 1.73× above sharp.**\n\nRemaining slack of the best *formula-free but computable* bound against sharp:\n1.00× at @19, where Y ≤ 19 is already the whole mid set, 1.22× at @23, 1.73× at\n@29. Remaining slack of the\nsharp constant against the strongest lower bound available, ρ·(g_max − 1): 5.3×\nat @19, 6.5× at @23, 8.7× at @29, widening.\n\n**Avenue (b), Erdős-Turán with the exact Fourier coefficients, loses on both\naxes.** ET(x) is *worse* than the standing lemma up to @17 (ratios 0.60, 0.68,\n0.83, 1.00) and only ×1.22 better at @19; its per-fold step is about 2.45 against\n3, so it crosses late, and it costs O(W) to evaluate against the blocked bound's\nN_Y·3^{k−|Y|}. This is the natal-object analogue of `level-ledger-tight.md`\n§4(c), which found the same shape on the twin tile and the same L¹ floor beneath\nthe whole family. Consistent with that file; it is not independent evidence.\n\n**Growth, measured, not derived.** D_x steps by 1.874, 1.946, 1.605, 2.261,\n1.698, 1.621 over @7 → @29, geometric mean 1.8208 against √3 = 1.7321. That sits\nabove √3 in the same direction and by about the same margin as\n`level-ledger-tight.md`'s R\\*(19) surprise, and for the same unseparated reason:\nD_x is a maximum, and no closed form for it is offered here.\n\n**A measured regularity with no proof attached.** sup|G| / D_x reads 0.5106,\n0.5046, 0.5020, 0.5011, 0.5004, 0.5002, 0.5001 at @7..@29, so the anchored\n(one-endpoint) channel is half the two-endpoint one to four digits and\ntightening. G is antisymmetric about its own centre to that precision. No\nderivation is offered and none was attempted; flagged as measured only.\n\n## 5. What the tighter constant buys the certificate engine\n\n`natal-cap-28-analytic-certificate.js` RESULT 1 certifies a prime q, of index\nj = idx(q), when 2^{j+1}·(2·3^k + 1) ≤ ½·(main − s). The 2^{j+1} is two sides\ntimes 2^j divisors d | P_j, and each term is a count of the comb dilated by\n(q·d)^{−1}. Replacing the universal 2·3^k by a per-dilation blocked bound gives\n\n | cap₂(q) − main | ≤ Σ_{sides} Σ_{d | P_j} ( B(α_{side,q,d}) + 1 ),\n\nwhich is the same theorem with a computed constant per term. Running cap-28's own\ncertification test against it, with Σcap₂ taken from cap-28's embedded K0 anchors\nand the old counts and shares asserted to reproduce:\n\n| level | certified primes, old | share of Σcap₂, main-term mass, old | certified primes, new | share, main-term mass, new | mean per-term constant, max over certified primes |\n|---|---|---|---|---|---|\n| @17 | 1 | 9.69% | **4** | **28.14%** | 17.4 against 2·3^4 = 162 |\n| @19 | 3 | 17.37% | **6** | **27.97%** | 47.9 against 2·3^5 = 486 |\n| @23 | 6 | 22.88% | **9** | **29.40%** | 112.1 against 2·3^6 = 1458 |\n\nBoth column labels say what the producer computes, and neither is the thing its\nold label named. The share column accumulates `cap = main + s` with\n`main = (A+B)·NW·Pj`, the analytic main term, not the exact cap₂: on exact cap₂\nmass the new share at @17 is **28.21%**, and the old column is 9.693% either way,\nwhich is why the custody assert against cap-28's 9.69% passes and hides the\ndifference (`redteam-0828-engine.md` §6.4, exact cap₂ counts). The constant\ncolumn is `errNew/2^{j+1} − 1`, an average over the `2^{j+1}` terms, so it is the\nmaximum over certified primes of the *mean* per-dilation constant; at @17 that\nmaximum is 17.41, at `q = 23`, while the maximum single-dilation range is 18.369,\nat `q = 31`. Both are far below `2·3^4 = 162` and the reading is unchanged.\n\nThe old shares reproduce cap-28's embedded 9.69 / 17.40 / 22.93% to 0.05\npercentage points, which is the custody check on the main-term arithmetic; the\nold counts 1 / 3 / 6 are asserted exactly.\n\nRead this at the right register. The certified head quadruples in count at @17,\nfrom 1 to 4, doubles at @19, and moves from 22.88% to 29.40% at @23. cap-28's own next-step\n(iii) predicted that a proven sub-3^k bound would push the certified share \"past\n50% at @29+\"; at the three levels reachable here it does not reach 50%, and the\ngain at @23 is the smallest of the three. The counts are floors: at @19 and @23\nthe per-dilation constant used is the blocked bound with Y ≤ 13, above the sharp\nper-dilation range, and at @17 alone the two coincide. Nothing here touches the\nengine's two unproven ingredients, so the engine's outputs keep the status of\ntheir weakest input, which remains the Buchstab Transfer Hypothesis, HEURISTIC.\n\n## 6. Status of item 8(c)\n\n**Partly answered, and not in the channel that pays.** For the α = 1 window\nchannel the constant is now exact at @7..@29 and computable at any level, with a\ngain over 2·3^k that grows rather than staying flat. For the dilation-uniform\nchannel, which is the one the certificate engine consumes, the best proven bound\nis still `level-ledger-tight.md` §6c's 3^{k−4}·26.682612, a constant gain of\n6.07, and the construction here does not improve it. What the engine actually\ngains, §5, comes from pricing each of its own finitely many dilations rather than\nfrom a better uniform constant, and that is a computation at each level, not a\nformula.\n\n## Defects noticed in passing\n\n- `certificate-engine.md`'s status table gives the Comb Discrepancy Lemma the\n scope \"every Legendre term is off its share by at most 2·3^k\". Each term is off\n by less than 1; the sum of the 2·3^k terms is off by at most 2·3^k. Wording,\n one cell, no downstream number affected.\n- `natal-cap-28-analytic-certificate.js`'s embedded OUTPUT at @13 prints\n \"agg NaN%\" for the certificate line, a 0/0 when cert.n = 0. Cosmetic.\n\n## What would falsify this, and whether that check has run\n\n- **Theorem A is false if some window beats D_x.** The check is the assertion\n that cap-25's six embedded grid maxima all sit at or below D_x. RUN, passes at\n all six.\n- **Theorem B is false if the blocked bound ever falls below D_x.** The check is\n the identity blocked(Y = all mids) = D_x, since a Y that keeps everything has\n the single block d = 1. RUN and asserted to 1e−9 at @7..@23; not run at @29,\n where W_Y = W is not enumerable by that path.\n- **The census could be wrong, making every D_x wrong.** The check is that the\n ascending pass counts exactly N = 2·∏(p−2) members. RUN, asserted at all seven\n levels.\n- **§5's arithmetic could be wrong.** The checks are that the old certified\n counts reproduce cap-28's 1 / 3 / 6 exactly and the old shares reproduce\n 9.69 / 17.40 / 22.93% within 0.6 points. RUN, both pass.\n- **The new counts in §5 could be capped rather than real.** The check is whether\n the divisor walk stopped on a certification failure or on the search cap\n JCAP. RUN: it stopped on a failure at all three levels, so JCAP was not\n binding.\n- **Unchecked, and flagged as such.** No independent re-derivation of D_x by a\n second method exists at @29; the α = 1 pass is the only witness there. At\n @7..@23 a second, independently written pass reproduces every D_x, the census\n and the sup|G|/D_x column (`redteam-0828-engine.md` §6.1), and it also\n reproduces the §5 @17 row with exact cap₂ counts and sharp per-dilation ranges\n (§6.4 there); @29 was outside that pass's compute budget too. The\n sup|G| / D_x → ½ regularity of §4 has no proof and no control. The per-dilation\n §5 constants are not compared against the sharp per-dilation ranges at @19 and\n @23, so the true engine gain at those two levels is larger than reported by an\n unmeasured amount.\n", "research/history/staging/engine-0830-at43-bigint.md": "# The @43 arithmetic ceiling: which paths leave 2^53, a BigInt-promoted K-30 engine gated against the record, and the price of the sixth residual collapse\n\n\n\n*(2026-08-30. Staging note; nothing here is integrated into a live document.\nHELD. Producer, formally embedded with both streams:\n`research/history/staging/engine-0830-at43-bigint.js`. Calibration marked per\nclaim: PROVEN, VERIFIED (exact computation), MEASURED, HEURISTIC, OPEN,\nREFUTED. Every figure below sits in that producer's OUTPUT block or is quoted\nfrom `research/natal-cap-37-at41-march.js`'s embedded OUTPUT by line; nothing\nis transcribed by hand. This note edits no other file.)*\n\n---\n\n## 0. What is open first\n\n- **beta(43) is not measured.** Nothing below scores the forecast on record.\n The producer can run @43 (modes `shard` and `combine`, positions in BigInt)\n and was not asked to, because the measured price is far above the 3.5 h\n rule of the brief: 579 h wall on this machine's eight shards, MEASURED by\n extrapolation from a slice (section 4), against the record's 6.035 h for\n @41. The decision the brief wanted is therefore made by the price, not by\n a result.\n- **S(41) still has one witness.** The record says so\n (`natal-cap-37-at41-march.js` reading 5, 10(a)); this note reproduces 0.248%\n of the @41 tile on the new path and nothing more of it. A full second\n march of @41 is 6 h of eight cores and was not run.\n- **The per-cell cost at @43 is 2.23x the @41 cost and is not decomposed.**\n The cursor scan is the obvious suspect (3.12 scour cursor entries per\n block cell at @43 against 0.53 at @41, printed in PART 3) and the record's\n own reading names the ring-bucket sieve as the fix that was not built; the\n number here is a MEASURED ratio on interleaved equal spans under a shared,\n heavily loaded machine, not a profile.\n- **No box conversion rate exists on disk.** `bench/README.md` carries laptop\n reference numbers and no box result file; the plan in section 6 prices in\n laptop core-hours and says so.\n\n## 1. The object, the engine, and the forecast on record\n\nThe object is beta(x) = S(x)/E(x) of the natal march: S(x) is the exact\nnumber of positions r = 11, 17 (mod 30) in [0, x#) surviving the strikes\nr = 0, -2 (mod p) for every prime 7 <= p <= x and every scour prime\nx < q <= floor sqrt(x#); E(x) is the independence value\n(2/30) x# prod(1 - 2/p) prod(1 - 2/q). The engine of record is the K-30\nengine in `research/natal-cap-37-at41-march.js`, whose header (lines 7-31)\nstates the question, the zero-knob classical curve\nbeta = (e^{2 gamma}/4)(1 + 2/ln W + 6/ln^2 W), and the ceiling. Rung:\nevery beta on the ladder is MEASURED (an exact integer S over a double E).\n\nThe record's last point is beta(41) = 0.8455 at S(41) = 256,725,962,834,\nE(41) = 303,627,067,641.7 (`natal-cap-37-at41-march.js`:825), the fifth\nconsecutive collapse of the classical residual, +0.0046, +0.0026, +0.0016,\n+0.0010, +0.0006 at @23..@41 (line 847). The forecast the TODO item quotes\nis the record's own last printed line (line 851): \"forecast @43\n(W = 1.308e16, BEYOND 2^53 — needs a different engine): classical-raw 0.8393\n(+persisted residual 0.8399)\". PART 4 of the producer re-derives both from\nthe formula: classical(43) = 0.8393 at W = 43# = 13082761331670030, and\npersisted = classical(43) + residual(41) = 0.8399, with residual(41) =\n+0.0006 from S(41)/E(41) = 0.8455 against classical(41) = 0.8449. So the\n@43 level extends beta, the natal survival ratio, at the MEASURED rung, and\nthe forecast is a HEURISTIC (a fitted-free curve plus the last residual). No\nband was registered for either forecast, in the record or in TODO item 1;\nthe two differ by 0.0006, which PART 4 prints as 6.4 units of the fourth\ndecimal beta is quoted in.\n\n## 2. The overflowing path, isolated\n\nPART 0 recomputes the record's fourteen-path audit at @43 with every worst\ncase formed in BigInt. Five paths exceed 2^53 = 9007199254740992 in\nmagnitude: A (W = 43# = 1.4525 x 2^53), B (the bracket (sq+1)^2), C (the\ny-depth product q^2 with q = 114379879), F (the CRT anchor, worst\n13082756491294884), and M (the position 30k + 17, worst 13082761331670017).\nThe other nine stay below 2^53: M = W/30 = 436092044389001, N = 2 prod(p-2)\n= 232312693362750, the shard anchor kA mod q, the cursor kb + i, and the\nInt32 paths (block offset 118574183 and stride 114379879, 5.5% and 5.3% of\n2^31). The extended-Euclid path E is bounded by 2q, not q^2, because each\ncoefficient is bounded by the modulus; the record's audit listed q^2 as its\nworst case, which was safe at @41 and is over-conservative.\n\n**The brief's premise is wrong in a way that matters for @47 and not for\n@43.** TODO item 1 says \"one line blocks it: W = 1.308e16 > 2^53 in the CRT\nanchoring product\", following the record's line 683 (\"@43 dies at path A\nfirst\"). PART 0 re-runs the record's path F in the record's own double\narithmetic, `(((a-b) mod q) * inv30(q)) mod q`, on all 26,157,448 scour\nclasses at @43 (both lanes): 644256 of the products exceed 2^53 in\nmagnitude, and 0 disagree with BigInt. The reason is parity: a - b is q - 11, q - 13, q - 17 or q - 19,\neven for every odd q, so each product is an even integer below\n2^54 = 18014398509481984 and is exactly representable in a double. The same\nparity makes W = 43# (even, below 2^54) exact as a double, which the block\nconfirms (\"double(W) exact: YES\"). The record's path F is exact at @43 by\naccident, and the accident ends at @47, where 47# = 614889782588491410 lies\nabove 2^54. What actually breaks the record engine at @43 is (i) its own\nguard `assert(W < 2**53)` in `levelSpec`, and (ii) path M: an odd position\n30k + 17 above 2^53 rounds to an even double, and PART 0 shows the first\nsuch k = 300239975158033 (BigInt 9007199254741007, double 9007199254741008,\noff by -1), over 31.2% of the @43 tile, so every sampled survivor there would fail its twin check. VERIFIED\n(exact computation, printed). The correction to the record is listed in\nsection 7.\n\n## 3. The BigInt path and its chain of custody\n\nThe producer copies the record's march kernel (wheel prefix, strike loop,\nSWAR census, survivor sampling) unchanged, and promotes exactly the five\npaths above: W and the primorial ladder are BigInt (the double copy is\nchecked exact and its exactness printed); the square-root bracket is a\nBigInt integer root; the y-depth assert is BigInt; the CRT anchor is a\nBigInt product mod q whose result, below q < 2^31, is stored as a double\nfor the cursor arithmetic; every position that leaves the march is a\nBigInt (`30n k + b`), and the 13-base Miller-Rabin plus strong-Lucas layer\nalready accepts BigInt. The class tables at @43 are 26,157,492 entries\nbuilt in BigInt in 6.41 s on the stderr line, a one-off per shard and not\na march cost. The march's per-strike cost is untouched: the\nrecord's PART 0.5 measured a blanket BigInt cursor at 30x slower, and this\npath keeps the cursors Float64 because at @43 they stay below 2^53 (paths G,\nH).\n\nCustody, PART 1, all on this code path, all VERIFIED against the record:\n\n- Gate 0: @7..@29 serial. S, N, E, beta, scour census and y match at all\n seven levels.\n- Gate 1: @31 serial. S(31) = 283449187, N = 4151035350, E = 328601798.6,\n beta = 0.8626, scour 37534 (37..447829), first survivor 448157, last\n 200560488827; 36 sampled survivors are twins.\n- Gate 2: @37 through eight shards, and per shard: all eight (np, S) pairs\n equal the record's shard lines (`natal-cap-37-at41-march.js`:784-791),\n S(37) = 7998394865, N = 145286237250, E = 9377228928.8, beta = 0.8530,\n first 2725001, last 7420738134527; 84 sampled survivors are twins. The\n eight shards took 1003.4 s wall here against the record's 437.0 s (line\n 795), the machine being shared with the other agents of the wave.\n- Gate 3: a slice of @41, shard 0's first 6000 blocks of 302246 at blockK\n 2^22, against the three progress ticks the record embeds (lines 803-805):\n blk 2000 np=4686722957 S=349203305, blk 4000 np=9373445779 S=661452618,\n blk 6000 np=14060168660 S=961752168, all three matched, and the first five\n @41 survivors (17442827, 17443091, 17443301, 17443661, 17443961; line 828)\n matched. This is 0.248% of the @41 tile.\n\nThe brief asked for @37 and @41 \"digit for digit\". @37 is met in full. @41\nis met on the slice the record makes checkable; the whole @41 tile is 6.035 h\nof eight cores (line 820) and is outside this note's budget. What the slice\nestablishes is that the promoted path is the same deterministic function of\nk as the record's on the region tested; it does not add a second witness to\nS(41).\n\n## 4. The price of @43, measured on a slice\n\nPART 3 prints the exact strike budgets 4 M sum 1/q: @41 5.9157e+13 (M =\n10141675450907, 1117909 scour primes, sum 1/q = 1.458276) and @43 2.6896e+15\n(M = 436092044389001, 6539362 scour primes, sum 1/q = 1.541874). The ratio\nis 45.465, of which 43.000 is the tile and the rest the longer scour sum.\n**TODO item 1's \"~13x the @41 runtime\" is not the strike budget and is not\nthe measured cost.** No derivation of 13 was found in the record; it is\nrecorded as a brief error in section 7.\n\nThe measured part: equal spans of 2^27 k-cells at k = M/3 were marched at\n@41 and @43 back to back, three interleaved repetitions, blockK 2^22. Per\nspan the block prints the counts (np and S on the span at each level) on\nstdout and the seconds on stderr: @41 2.33, 2.19, 2.17 s and @43 5.24,\n4.89, 4.87 s, medians 2.19 s and 4.89 s, a per-cell cost ratio @43/@41 of\n2.230 (per strike 2.109). The load average at probe time is printed beside\nit (7.6, 66.9, 124.9 on 10 cpus), since this machine was shared with the\nother agents of the wave. The extrapolation, printed on stderr, is 6.035 h\n(the record's eight-shard wall for @41) x 43.00 (tile) x 2.230 (per cell) =\n579 h wall = 24.1 days on this machine's eight shards; the brief's 13x would\nhave been 78 h. A single serial span at blockK 2^24 came out at 3.65 s\nagainst 4.89 s at 2^22, which is the same serial-versus-concurrent\ninversion the record's cost log found at @41 (line 742 and reading 3), so\nthe block size for a real run has to be re-probed under concurrency on the\nmachine that runs it.\n\nCalibration: MEASURED, one machine, loaded, three repetitions, a single\nspan position. The 2.23 factor is the least certain input; the 43x tile is\nexact. Even at a per-cell factor of 1.0 the run is 6.035 h x 43 = 260 h\nhere (arithmetic on printed values). Nothing\nunder this heading is under the 3.5 h rule.\n\nThe residual-collapse question priced: the sixth collapse is the whole\npayoff of the run, and it is one number to four decimals. The residual\nsequence +0.0046, +0.0026, +0.0016, +0.0010, +0.0006 has ratios near 0.6;\na sixth term of that size sits between the two forecasts, 0.8393 and\n0.8399, which differ by 0.0006. So the point would distinguish raw from\npersisted only if it lands near one of them, and a landing between them\nscores as the nearer one under the record's own rule (line 846,\n\"closest = ...\") without a registered band.\n\n## 5. What the @43 point would decide\n\nOne calibrated sentence: the @43 point would add a sixth MEASURED term to\nthe classical residual series and score the HEURISTIC forecast pair\n0.8393/0.8399, and it cannot move any proof, because beta is the knob whose\nheadroom \"has no consumer\" (TODO.md:562, `sift-limit-attack.md`:689) and\nthe fifth collapse already leaves the classical curve without a rival\ndescription at any computed level (`paper/anchored-note.md`:527-533). The\nrun's value is calibration of a curve, at weeks of box time.\n\n## 6. The box plan\n\nGiven only if Chris wants the point. Everything is script-only: the\nproducer is one file with no repo dependency; ferry it, run, ferry the JSON\nback, `combine` on the laptop.\n\n1. **Convert first.** Run `bench/bench.sh` on the box for the Node\n single-core rate, and this producer's `probe 43 22 32`, `probe 43 23 32`,\n `probe 43 24 32` as 16 concurrent copies each (a shell loop), to pick the\n block size under concurrency on the 5950X's two 32 MB CCDs. The record's\n @41 finding was that the serial optimum (2^24) loses under eight shards;\n the box may differ. The producer's `BLOCKK` is a constant and must be\n edited on the ferried copy if 2^22 loses; say so in the returned log.\n2. **Segment.** `shard 43 i 128 ` for i = 0..127, 16 at a time under\n `nohup nice`, each writing `e0830-x43-of128-i.json` with np, S, first and\n last survivors and samples as decimal strings, plus its own seconds and\n the box's node version and CPU model. A segment is the checkpoint: a\n crash loses at most one segment. At the laptop rate each segment is\n 579 h x 8 / 128 = 36 core-hours (arithmetic on the printed 579 h); at 16 concurrent that is 8 rounds of\n 36 h = 12 days if the box matches this machine per core, and unknown\n until step 1 has run. Memory per shard is about 0.6 GB (26M-entry class\n tables plus cursors), 16 shards under 10 GB.\n3. **Fold and score.** `combine 43 128 ` on the laptop: asserts the\n partition is contiguous and exact, N = 2 prod(p-2), sums np and S,\n cross-checks E against the 2^256 fixed-point reference, BPSW-checks every\n sampled survivor as a BigInt, and prints the misses against 0.8393 and\n 0.8399 with the sign. Bind the box stdout/stderr by the declared-provenance\n path with the node version and CPU in the tail; never hand-paste.\n4. **Custody on return.** Before trusting S(43): re-run `quick` and the\n gates on the box's copy of the file (the `full` mode does this; its\n embedded elapsed here is 1621.4 s on the loaded laptop), and compare a segment boundary against a laptop\n re-run of the same k-range (`shard 43 i 128` for one i, about 36\n core-hours, or a narrower range via `probe`).\n\n## 7. Corrections to the record and to the brief\n\n- **Brief and record, the blocking path.** TODO item 1 line 554 and\n `natal-cap-37-at41-march.js`:683 say the CRT anchoring product / path A is\n where @43 breaks. PART 0 shows path F and path A are exact at @43 by\n parity (0 disagreements on 26,157,448 classes; W even and below 2^54);\n the breaks are the engine's own guard and path M (survivor positions\n above 2^53, 31.2% of the tile). The brief's fix (BigInt on \"that single\n path\") would have been insufficient without path M and unnecessary for\n path F at @43; it becomes necessary for F at @47. The producer promotes\n all five.\n- **Brief, the cost factor.** \"~13x the @41 runtime\" has no source found in\n the record; the strike budget ratio is 45.465 and the measured per-cell\n extrapolation is 579 h against the record's 6.035 h, about 96x.\n- **Record, the audit's path E bound.** `natal-cap-37-at41-march.js` PART 0\n lists invMod's worst case as q x q; the extended-Euclid coefficients are\n bounded by the modulus, so the true worst case is 2q. Harmless at @41 and\n at @43.\n- **Brief, the reading of TODO's forecast.** The brief guessed the forecast\n was for \"the anchored bias beta or the anchored calm\"; it is beta = S/E of\n the natal march, and not the anchored calm of `research/anchored-calm.md`,\n which is a different object.\n\n## 8. Honesty\n\nThe engine claim is narrow and checked: the promoted path reproduces the\nrecord wherever the record can be checked without a six-hour run. It has not\nproduced a number the record does not already hold. The cost claim rests on\none loaded machine and three interleaved spans; it is good to perhaps a\nfactor of 1.5 on the per-cell factor and exact on the 43x, and no reading of\nit brings @43 under the run-it-now rule. The @43 forecast is unscored and\nstays so until someone spends the box time; this note's recommendation is\nthat the point be scheduled only if a sixth term of the residual series is\nwanted for its own sake, since no live target consumes it.\n", "research/history/staging/fekete-1d.md": "# TODO 1d: the bounded-defect Fekete route, probed at 47#\n\n\n\n*2026-08-20. Producers: `research/attack-fekete-1d-01-defect47.js` (0.1 s,\n`code-sha256 45554df4`, `out-sha256 afc6d190`, formal embed, seven numbered\nreadings) and `research/attack-fekete-1d-02-lemma.js` (0.2 s,\n`code-sha256 62cb37d9`, `out-sha256 1c99d42b`, formal embed, six numbered\nreadings). Both fingerprints verified with `embed.js --check` after writing.\nNothing was committed or pushed; no existing corpus document was edited.\nLegend as in `research/sift-limit-attack.md`: **[PROVEN]** published theorem or\na proof given here; **[VERIFIED]** checked computationally here; **[MEASURED]**\nempirical, finite range; **[INFERRED]** deduction from sourced facts;\n**[CERTIFIED]** follows from a replay-verified corpus certificate.*\n\n---\n\n## 0. The verdict, up front\n\n> **The defect at 47# is ordinary, the one unverified term cannot move any 1d\n> reading, and — the sharpest thing this pass found — the 47# enumeration\n> cannot move 1d's TPC threshold either, whatever value it returns. The\n> bounded-defect Fekete lemma is stated exactly, proved, and every hypothesis\n> except the candidate itself is now discharged. The route is reduced to one\n> named inequality.**\n>\n> 1. **The defect at 47#, exactly.** Ten integer pairs land in `[47, 53)`. The\n> largest defect is `D = 1.0330` nats at `(4,12)` — second largest on the\n> whole 111-pair table, 0.0431 under the global sup `1.0761` at `(4,10)` —\n> and the all-prime pair `(7,7)` reads `−0.2400`. The 47-window ranks 3rd of\n> 21 windows. Nothing about it is extreme. **[VERIFIED]**\n> 2. **The 704-vs-708 discrepancy is immaterial.** The greedy's replay-verified\n> covering gives `G₂(47#) ≥ 705`; A144311 says 708. The worst shift any\n> defect reading can take from that bracket is `ln(708/705) = 0.00425`\n> nats — 66× below the smallest margin in play. A144311 would have to be low\n> by ≥ 32 (4.5%) before the reachable sup C moved. **[VERIFIED]**\n> 3. **REFUTATION: 1c's \"each new ladder-known x raises 1d's TPC threshold\" is\n> false at 47#, and certified false — no enumeration outcome can change\n> it.** The custody integer threshold is `S(16) = ln(256/66) = 1.3555`, a\n> 13#-fact. The 47-window's best is `S(52) = ln(2704/G₂(47#)) ≤\n> ln(2704/705) = 1.3443 < 1.3555`, capped from above by the certified\n> greedy floor. Same freeze at 53# (cap 1.3524) and 59# (cap 1.3280). The\n> first term that can raise it is x = 61, and a 61# enumeration is\n> ~146,969× the 43# run. **[CERTIFIED]**\n> 4. **The defect reads BOUNDED on a second instrument, and custody-only.**\n> Per-window sup slopes: G₂ minus the constant-defect null =\n> `−0.0414 ± 0.0846` (22 terms) and `−0.0188 ± 0.1466` (14 custody terms) —\n> consistent with bounded both times, and the verdict does not lean on the\n> eight literature terms. **[MEASURED]**\n> 5. **What a bounded defect buys, exactly.** The multiplicative Fekete lemma\n> (stated and proved in `attack-fekete-1d-02-lemma.js`, eight lines):\n> conditional on the candidate with ANY constant, `β = lim ln G₂(x#)/ln x`\n> **exists** and equals `inf_n (ln Ĝ(n) + K)/ln n`; and\n> `β < 2 ⟺ S(n) > K at a single integer`. Existence is the honest prize;\n> the explicit-constant version is TPC-strength. **[PROVEN, given the\n> candidate]**\n\n---\n\n## 1. What \"probe 47#\" can and cannot mean, priced\n\nThe exact enumeration at 47# is already priced by direct probe: **1.49 days per\nrun, two disjoint-mask runs required** (`phase1-T2b-exact-ladder.md`), so it was\nnot launched here — that is 1c's task, with 1c's Poisson-window\npre-registration discipline. What the corpus already holds at 47# is a bracket:\n\n| source | value | grade |\n|---|---|---|\n| greedy covering of `[1, 704]`, replay-verified | `G₂(47#) ≥ 705` | CERTIFIED (`greedy-oracle-validation.md` §3) |\n| OEIS A144311 a(15), Alekseyev 2009 | `G₂(47#) = 708` | literature, unverified here |\n\nEvery reading below is taken at both endpoints of that bracket, which is what\nmakes this a probe rather than a wait.\n\n## 2. The defect at the 47-window\n\n`D(s,t) = ln[Ĝ(st)/(Ĝ(s)Ĝ(t))] = S(s)+S(t)−S(st)` (the identity, PROVEN in\n`import-interp.md` §3, re-verified here to `5.6e−16` at all pairs). The ten\npairs with `st ∈ [47, 53)`, at the published 708:\n\n| pair | st | R | D (nats) |\n|---|---|---|---|\n| (4,12) | 48 | 2.8095 | **1.0330** |\n| (5,10), (6,8) | 50, 48 | 1.9667 | 0.6763 |\n| (3,16), (4,13) | 48, 52 | 1.7879 | 0.5810 |\n| (2,24), (2,25), (2,26) | 48, 50, 52 | 1.7353 | 0.5512 |\n| (3,17) | 51 | 1.0926 | 0.0886 |\n| (7,7) | 49 | 0.7867 | −0.2400 |\n\nThe window's sup ranks **3rd of 21 windows** (behind p = 37's 1.0761, which is\nthe global sup, and p = 71's is 1.0202 just below); the running per-window sup\nlast rises at the p = 37 window and is flat over the ten windows after it.\n**[VERIFIED]**\n\n**Sensitivity.** At `G₂(47#) = 705` and at 708 alike: sup C stays 2.9333 at\n(4,10), every threshold stays put, and the maximal shift of any D is 0.00425\nnats. The flip value is `G₂(47#) ≥ 740` — pair (4,12) would then overtake\n(4,10) — i.e. A144311 low by ≥ 32. Below 705 is excluded by the certificate.\n**[VERIFIED]**\n\n## 3. The threshold freeze, which corrects two records\n\nTPC-implication for an explicit constant is `ln C < S(b)` at a ladder-known\ninteger base `b` (`import-interp.md` §4). The threshold column, recomputed:\n\n| ladder | threshold `max_b S(b)` | at base |\n|---|---|---|\n| full 22-term (8 literature terms) | 1.3946 | b = 66 — **rests on Wang 2024's a(18), `G₂(61#) = 1080`** |\n| custody-only, 14 exact terms | **1.3555** | b = 16 — rests on `G₂(13#) = 66` |\n| custody + 47# verified (at 708 or 705) | 1.3555 | b = 16, unchanged |\n\nBecause `S(b) = ln(b²/Ĝ(b))` **falls** as `Ĝ` rises, a certified lower bound on\n`G₂` is a certified **upper** bound on the window's best S. The greedy floors\ntherefore freeze the next three windows below the standing custody max:\n\n| window | best base | cap `ln(b²/floor)` | can raise 1.3555? |\n|---|---|---|---|\n| [47,53) | 52 | 1.3443 | no, certified |\n| [53,59) | 58 | 1.3524 | no, certified |\n| [59,61) | 60 | 1.3280 | no, certified |\n| [61,67) | 66 | 1.3992 | **yes** — fires iff verified `G₂(61#) ≤ 1123` (published: 1080) |\n\n**Two corrections follow.** (i) TODO 1c's stated win \"each new ladder-known x\nraises 1d's TPC threshold, directly loosening the constraint\" is wrong at 47#\n— and not merely at the published value: **certified wrong for every possible\nenumeration outcome**, and equally at 53# and 59#. The first riser is x = 61,\nwhose enumeration is `47·53·59 = 146,969×` the 43# run (~1 h on ten cores) —\nout of reach by ~5 orders of magnitude. (ii) The corpus's quoted integer\nthreshold 1.3946 is a **literature-conditional** number (Wang 2024's a(18));\nthe custody-grade threshold is 1.3555 and is effectively frozen there.\n**[CERTIFIED / VERIFIED]**\n\n**The 1d position now stands on custody terms alone.** The reachable sup\n`C = 2.9333` is attained at (4,10) from `G₂(7#) = 30` and `G₂(37#) = 528`, both\ncorpus-exact; the threshold 1.3555 is a 13#-fact. The candidate's TPC window is\n`C ∈ [2.9333, 3.8788)`, `ln C ∈ [1.0761, 1.3555)` — 0.2794 nats wide — with no\nliterature term anywhere in it. Any proof of the candidate with a constant in\nthat window is a proof of `β < 2`. **[VERIFIED]**\n\n**One domain-cap artifact repaired in passing:** the recorded integer floor-7\nsup 1.5533 (`import-interp.md` §4) was taken under `st ≤ 79`; `Ĝ` is known on\n`[2, 83)`, and the widest legal domain admits (8,10), st = 80, with\n`R = 1.9000` — equal to the recorded real-form sup, so the real/integer split\nat floor 7 was the cap, not structure. No verdict changes. **[VERIFIED]**\n\n## 4. Is the defect bounded?\n\n**Second instrument, same verdict as `import-interp.md` §5.** Per-window sup\n`Wmax(p) = max{D(s,t) : st ∈ [p, p_next)}`, slope against `ln p`, nulls sampled\nat the same primes and stepped the same way in the same pass:\n\n| ladder | slope ± se | G₂ − POW contrast |\n|---|---|---|\n| G₂ full (22) | +0.1661 ± 0.0637 | **−0.0414 ± 0.0846 → consistent with bounded** |\n| POW null, constant defect | +0.2075 ± 0.0558 | (the calibration: this IS what bounded looks like here) |\n| LOG null, decreasing defect | −0.2415 ± 0.0554 | (the instrument is not pinned positive) |\n| G₂ custody-only (14) | +0.2526 ± 0.1094 | **−0.0188 ± 0.1466 → consistent with bounded** |\n\nG₂ reads *less* growing than the constant-defect null on both ranges, and the\ncustody-only row shows the bounded verdict does not lean on the eight\nliterature terms. **[MEASURED]**\n\n**The instrument's blind spot, stated as a limit.** Every power-log law\n`Ĝ ~ c·x^β (ln x)^δ` **with δ ≥ 0** has a bounded window-sup defect\n(continuous limit `δ·ln(1/ln 2) − ln c`; for δ < 0 the k = 1 rung diverges\nand the defect is unbounded, sign lemma PROVEN in\n`attack-0829n-hsubpow-K.md` §3b, qualifier added 2026-08-29), because the\nsup is driven by small-s pairs; an\nunbounded defect would live on the diagonal, where the largest `√x` any ladder\noffers is ~9. So \"bounded\" is the supported reading under every law the corpus\nentertains, and is **not** a demonstrated fact about the true defect — which is\nexactly why the honest Fekete target carries an unnamed constant.\n**[INFERRED]**\n\n## 5. The exact lemma, hypotheses discharged\n\n> **Lemma (multiplicative Fekete with bounded defect).** Let\n> `f : {2,3,4,…} → [0,∞)` satisfy **(H-sub)** `f(st) ≤ f(s) + f(t) + K` for all\n> integers `s,t ≥ 2` and some constant `K ≥ 0`, and **(H-mono)** `f`\n> nondecreasing. Then `lim f(n)/ln n` exists and equals\n> `L = inf_{n≥2} (f(n)+K)/ln n ∈ [0, ∞)`.\n\nProved in eight lines in `attack-fekete-1d-02-lemma.js`'s header (liminf from\nthe definition of inf; limsup by iterating (H-sub) on powers of an arbitrary\nbase and bridging with (H-mono)). de Bruijn–Erdős Theorem 22 — the\ngrowing-defect version — is **not needed**: the measured defect reads bounded\nand plain-Fekete shape suffices, exactly as `import-interp.md` §6 anticipated.\n**[PROVEN]**\n\n**Why BGT's obstruction does not apply here:** the lemma never asks an integer\nto factor. BGT's machine died on H1 (`π(st) ≠ π(s)+π(t)`, the ground set does\nnot split); this lemma uses (H-sub) on products only, and (H-mono) replaces the\nadditive ground set in the between-powers bridge. **[INFERRED]**\n\nHypotheses, for `f = ln Ĝ`:\n\n| hypothesis | status |\n|---|---|\n| `f ≥ 0` | trivial, `Ĝ ≥ 2` |\n| (H-mono) | **PROVEN one line + VERIFIED seven folds**: the twin-slot set of `T_next` is a subset of the periodized slot set of `T_x` (folding only deletes slots; deletion only merges gaps, so `G₂` is nondecreasing). Checked literally by direct sieve at every fold `2→3→…→19`, periods to `9,699,690 = 19#`, reproducing the exact ladder term by term; monotone at all 22 ladder terms |\n| (H-sub) | **OPEN — the candidate itself.** Holds at all reachable pairs with `K = 1.0761`; unproven beyond; the route's entire remaining content |\n\n**The conclusion formula, calibrated on known-truth controls** (campaign rule:\nestimator on a control in the same pass). On `POW` (true β = 1.546, constant\ndefect) the finite-range inf reads 1.5901 at n = 10⁶, +2.9% and falling from\nabove; on `LOG` (true β = 1, δ = 2) it reads 1.5335, **+53%**, with its K\ndriven by the (2,2) pair. Both approach from above, as the inf-over-a-subset\nargument requires. **So the lemma's value output is unquotable at any reachable\nrange; what it delivers is existence.** **[MEASURED]**\n\n## 6. What a bounded defect buys, and the two faces\n\nConditional on (H-sub) with **some** constant, never named:\n\n- `β = lim ln G₂(x#)/ln x` **exists** — along integers, hence along reals,\n since `Ĝ` is a step function. 1d's central question becomes well-posed:\n `β < 2` forces `G₂(x#)/x² → 0`, `β > 2` forces it to ∞, and only `β = 2`\n leaves the sublinear part to decide. **[PROVEN, given (H-sub)]**\n- `β < 2 ⟺ S(n) > K` at a single integer. This one equivalence is the trap and\n the prize at once: an explicit `K < 1.3555` is TPC-implying (no soft proof\n should be expected), an unnamed `K` buys existence without touching TPC. The\n corpus's `1.8992` at base 16 (`import-interp.md` §4) is this formula at the\n reachable K, reproduced exactly. **[PROVEN given (H-sub); cross-check\n VERIFIED]**\n\nNot bought: any value of β, any movement on `4.2665 → 2`, any unconditional\nstatement.\n\n## 7. NOT REACHED\n\n- **The 47# enumeration itself.** Priced at 1.49 days per run, two runs\n (`phase1-T2b-exact-ladder.md`); not launched. The bracket `[705, ∞) ∩\n {published 708}` is what this pass used, and §2-§3 show nothing 1d needs\n from the enumeration beyond what the bracket already certifies.\n- **No proof of (H-sub), no counterexample, no bound on the true K.** The\n pairs that would decide it need both arguments large; the ladder has none\n (largest `√x` ≈ 9).\n- **No exclusion of a growing defect.** §4's verdict is consistency with\n bounded, on an instrument blind to diagonal growth; a non-power-log law\n could hide there.\n- **de Bruijn–Erdős Theorem 22 remains unopened** (as in `import-interp.md`\n §9); it is not needed for the bounded shape, and nothing here cites its\n content.\n- **The real-base form of the candidate** was not re-examined; the corpus\n instruction to state the candidate on integer bases stands\n (`import-interp.md` §4).\n- **Whether 61# could ever be reached by a smarter engine** than direct\n enumeration (the only riser base in reach of the freeze analysis) was not\n investigated; at direct-enumeration scaling it is ~146,969× the 43# hour.\n\n## 8. Sources and reproduction\n\n| artifact | what it carries |\n|---|---|\n| `research/attack-fekete-1d-01-defect47.js` | the 47-window pairs, sensitivity at 705/708, flip values, the threshold freeze, the per-window boundedness instrument with nulls (§§2-4) |\n| `research/attack-fekete-1d-02-lemma.js` | the lemma with proof, (H-mono) verified at seven folds, the control calibration, the conditional application (§§5-6) |\n| `research/exact-g2-ladder.js` | the 14 custody terms, parsed at run time |\n| `research/import-interp-01-bgt-defect.js` | the corpus's A144311 copy, parsed at run time; the headline numbers cross-checked here |\n| `research/history/staging/greedy-oracle-validation.md` §3 | the replay-verified greedy floors 705/870/954/1075 |\n| `research/history/staging/phase1-T2b-exact-ladder.md` | the 47# price: 1.49 days per run |\n| `research/history/staging/import-interp.md` §§3-7 | the identity, the trap statement, the slice-instrument boundedness reading this pass confirms |\n\nReproduce with `node research/qc/embed.js --check\nresearch/attack-fekete-1d-01-defect47.js` and `node research/qc/embed.js\n--check research/attack-fekete-1d-02-lemma.js`; both fingerprints match as of\n2026-08-20.\n", "research/history/staging/item-x-offset.md": "# The ~3.8 law's −0.45% offset: the candidate family priced, two mechanisms tested, and the @37 discriminator\n\n\n\n*(2026-08-20. TODO item X, after both blind tests: `1 − J = 4S₂` is the last\nlaw standing and is NOT exact — a relative offset of −0.41% (@29) and −0.48%\n(@31), the same absolute residual ≈ 1.19e−4 at both sharp levels, detected at\nz = −2.32 on the corrected slot-clustered σ (@31) and only consistent at −0.43\n(@29). The open analytic target must now produce the offset, not just the law.\nThis file prices the finite-level-correction family against the exact\nresiduals, measures where the deficit lives, and pre-registers the @37 test\nthat separates the survivors. Producers, all three formally embedded:\n`research/attack-x-offset-01-terms.js` (170 s),\n`research/attack-x-offset-02-profile.js` (2948 s),\n`research/attack-x-offset-03-cofactor.js` (117 s). Pre-registration for @37:\n[xchan-at37-offset-prereg.md](xchan-at37-offset-prereg.md), committed alone.\nCalibration is marked on every claim: MEASURED, DERIVED, REFUTED, POST HOC.)*\n\n## Verdict\n\n**The offset's best available description is a finite-level correction of S₃'s\nsize and sign — `1 − J = 4S₂ − S₃` scores z = +0.04 (@29) and −0.41 (@31)\nwhere the standing law scores −0.43 and −2.32 — but it is POST HOC, and the\nblind test that decides it is @37, pre-registered and committed alone**\n([xchan-at37-offset-prereg.md](xchan-at37-offset-prereg.md), commit `0a7dd73`).\nWhat this pass settles outright, at high significance:\n\n1. **The σ behind the @31 detection is calibrated.** Seven known-truth control\n draws land inside 1.9σ; the instrument does not manufacture a −0.45%-scale\n offset (§4).\n2. **Two candidate corrections and one shape are REFUTED on existing data**:\n `4S₂ − 4S₃` (5.3σ), `4Σ1/(q(q+2))` (12.4σ), and the locality shape\n `1−J ∝ E_super[Σ_{q∈Q} q⁻²]`, whose constant would have to grow 120 → 572\n (§2).\n3. **The marginal-discrepancy mechanism does not carry the offset**: re-basing\n the denominator on measured per-prime strike counts moves the residual by\n +118% at @29 but by −55% at @31 (and −145%/−224% at the noisy levels) —\n sign-erratic where a real mechanism must be stable (§5).\n4. **The raw 11/17 line is itself biased**, J_line − 1 = +4.03e−3 at @31\n (z = 147.6), positive and decaying across four levels: the natal J that\n `4S₂` fits to half a percent is the NET of a positive lattice bias and a\n deeper sieve deficit, 1 − J/J_line = 1.15–1.18 × 4S₂ (§3).\n5. **The far-super-W surplus component is now DERIVED**: on the strips where\n the scour pair's only eligible cofactor is 1, the natal count is\n pairs·(1/4)·∏_{7≤p≤x}(p−2)/(p−1) — verified to 0.001% over 5.05e8 pairs at\n @31 — an enrichment over the smooth model of exactly ∏_{p≤x} p/(p−1)\n → e^γ ln x (§6b).\n6. **The deficit is a structured profile, not a constant** — octave deficits\n spanning 0.006 to 0.059 with sign flips, each bin hundreds of σ from the\n aggregate (§6a). Any derivation of the ~3.8 constant, and of its offset,\n must produce this profile; an aggregate-only argument cannot be checked\n against it and should not be believed.\n\n## 1. Units, and what the offset is\n\nEverything here is in J-units: `J = obs/CRT` is a dimensionless ratio of\ncounts, `obs` the mixed super-`W` triple census over the natal set of `[0,W)`\nand `CRT` the smooth model's expectation `6·N̄·Σ_{∏Q>W} 1/∏Q`\n(`research/xchan-at29-01-segmented.js`, whose embedded OUTPUT is the source of\nevery measured input below). Three different numbers describe the same\nmismatch and are not interchangeable:\n\n- **Δ = 4S₂ − (1−J)_meas**, the ABSOLUTE residual in J-units. Positive means\n the closed form predicts a deeper deficit than measured. Δ = **+1.201e−4**\n at @29 and **+1.185e−4** at @31 — the same number to 1.3%.\n- **d = (1−J)_meas/4S₂ − 1**, the RELATIVE offset, σ-free: **−0.41%** and\n **−0.48%**.\n- **z = −Δ/σ_slot**, the significance on the slot-clustered σ of the\n pooled-floor fix (`defect-repairs.md` item 3): **−0.43** (@29, a\n non-detection) and **−2.32** (@31, the one detection).\n\nSo the offset is one 2.3σ detection plus one consistent non-detection, and\nany model of it is a model of essentially one measured number. That is why\nthis file prices a FAMILY against exact arithmetic, tests mechanisms that are\nmeasurable at high significance today, and pushes the decision to @37.\n\n## 2. The candidate ladder against the exact residuals (producer 01)\n\n`research/attack-x-offset-01-terms.js` computes the correction sums exactly\n(S₃ = Σ_{x W, not about the sieve).\n\n## 5. The marginal-corrected denominator (producer 02, measurement 3)\n\nThe census accumulates the exact per-prime strike counts n_q^L, n_q^R over\nnatal slots and rebuilds the denominator from them:\n`CRT' = N̄·Σ_{i **#{natal r in the strip} = #pairs · (1/4)·∏_{7≤p≤x}(p−2)/(p−1)** —\n\nan enrichment over the smooth model of exactly **∏_{p≤x} p/(p−1) → e^γ ln x**\n(Mertens). **[MEASURED, DERIVED]** Verified per strip at four strips per\nlevel: aggregate relative deviation +0.191% (@19, n = 10,134), +0.026%\n(@23), +0.006% (@29), **+0.001% (@31, n = 68,409,827 natal points over\n5.05e8 pairs)**. Measured enrichment 6.5423 against derived 6.5423 at @31.\n\nThis is the mechanism of the far-octave and near-√W surpluses in §6a, on the\nline (enrichment 30/8 = 3.75, no base factors) and amplified on the natal\nset (∏ p/(p−1), growing like e^γ ln x) — the first exactly-derived component\nof the joint census. It is a SURPLUS component: it pushes J up, against the\nbulk deficit, and its weight is concentrated where the profile's sign flips.\nDeriving the full small-cofactor contribution (cofactors that are scour\nprimes and semiprimes, and the b-side divisor coupling) is the natural\ncontinuation of the item's analytic target, and it is a closed-form\ncomputation over prime sums, not a census.\n\n## 7. What this does not show\n\n- **No derivation of 4S₂, and none of the offset.** The candidate C1 = 4S₂ − S₃\n is a POST HOC scale match; the cofactor-layer law (§6b) derives a COMPONENT\n of the census, not the law. The open analytic target of TODO item X is\n narrowed, not met.\n- **The offset is still one detection.** Nothing here adds a sharp level; @37\n is where the family separates, and its pre-registration is committed.\n- **The fitted members are two-point descriptions** with ~40% parameter error\n bars; no shape claim is made from them.\n- **The multiplicative line/sieve split is a model.** J = J_line·(J/J_line) is\n arithmetic; reading J/J_line as \"the sieve-only deficit\" assumes the natal\n mask samples the line's lattice bias neutrally. The random-mask control\n supports that for random masks; the natal mask is not random.\n- **j = 4 remains untouched**, as in the record.\n\n## 8. Files touched\n\n- `research/attack-x-offset-01-terms.js`: new, embedded (170.2 s), READINGS\n appended below the OUTPUT banner, binding re-checked.\n- `research/attack-x-offset-02-profile.js`: new, embedded, READINGS appended,\n binding re-checked.\n- `research/attack-x-offset-03-cofactor.js`: new, embedded (116.8 s),\n READINGS appended, binding re-checked.\n- `research/history/staging/xchan-at37-offset-prereg.md`: new, committed\n **alone** before any @37 census of any kind exists.\n- `research/history/staging/item-x-offset.md`: this file.\n\nNo live document was edited. The TODO item X update is proposed in §10 and\nnot applied.\n\n## 9. Gate\n\n**Before:** `node research/qc.js` TOTAL = 2, both pre-existing and neither\nthis session's: the documented `natal-cap-27-t4-at13.js` embeds finding, and\nan `uncited-script` on another live session's `attack-advmin-1113.js`.\n\n**After:** `node research/qc.js` TOTAL = **0** — all eleven checks clean,\nincluding `scripts` (all three producers are cited by this file) and `embeds`\n(the three OUTPUT bindings verify with the READINGS appended). The two\npre-existing findings cleared under the shared tree while this session ran;\nnone of this session's files ever appeared in the finding set.\n\nAll three producers are formally embedded; READINGS were appended below the\nOUTPUT banners. `embed.js --check` re-verified code-sha256 and out-sha256 on\nproducers 01 and 03 by full re-execution; producer 02's binding is verified\nby the static `embeds` check above (its own `--check` is a second 2948 s\ncensus execution and was left running when this file was written — its\nverdict, PASS or FAIL, belongs to the session record, and a FAIL would be a\nreproducibility finding against a file whose two prior executions agreed).\n\n## 10. Proposed update to TODO item X (report only, not applied)\n\nReplace the clause *\"the derivation stays the open analytic target and must\nnow produce the offset too\"* with:\n\n> The offset now has a priced candidate family\n> (`research/history/staging/item-x-offset.md`): `4S₂ − S₃` fits both sharp\n> levels inside 0.5σ (POST HOC — found against the known residuals), the\n> ≥ 4S₃-sized corrections and the per-triple locality shape are REFUTED, the\n> marginal-discrepancy and instrument/σ mechanisms are closed by measurement,\n> and the @37 blind test that separates {no correction | shrinking S₃-scale |\n> constant absolute} at 4.8–9.1σ is pre-registered\n> (`xchan-at37-offset-prereg.md`, committed alone at `0a7dd73`; the @37\n> natal census is an overnight run on the existing segmented instrument).\n> Two structural facts now constrain the derivation itself: the natal J is\n> the net of a POSITIVE line-lattice bias (J_line − 1 = +4.03e−3 at @31,\n> z = 148, decaying) and a deeper sieve deficit (1.15–1.18 × 4S₂ under the\n> multiplicative split); and the deficit is a structured profile (octave\n> bins 0.006 to 0.059 with a sign flip, each 100σ+ from the aggregate) whose\n> far-super-W surplus component is exactly derived: the cofactor-one layers\n> are natal-enriched by ∏_{p≤x} p/(p−1) → e^γ ln x, verified to 0.001% at\n> @31 (`research/attack-x-offset-03-cofactor.js`). An aggregate-only\n> derivation of the ~3.8 constant is no longer checkable-complete: it must\n> reproduce the profile tables.\n", "research/history/staging/measure-0830-rho-sup-z41.md": "# The exact sup at z = 41: sup|ρ̃| over 7.42e12 positions and sup|R_H| at H = z³, z⁴, TODO item 0's first move (b), run\n\n\n\n> **RIDER 2026-08-30 (orchestrator, from `redteam-0830-rml.md`).** The\n> reflection identity ρ̃(W−3−y) = −ρ̃(y) is PROVEN under independent code,\n> together with the two hypotheses it needs (the three-entry COMB list closed\n> under swap with the weight fixed; the swapped class −2 − c mod q), verified\n> at every term of z = 13..37. The sealed forecast's arithmetic and the cost\n> extrapolation reproduce. Two custody limits, disclosed: the independent walk\n> reaches only z = 29, so sup|ρ̃|(31) = 28.122062, sup|ρ̃|(37) = 52.219092 and\n> sup|R_H3| at 31 and 37 are NOT re-derived (five of seven and three of five\n> is what the custody claim rests on after that pass); and the seal's TIMING\n> cannot be verified by any method available — every sealed number reproduces\n> from the seven cited sups alone, which is consistent with pre-registration\n> and is not evidence of when it was written (an out-sha256 binds content, not\n> the clock).\n\n*(2026-08-30, staging. Producer:\n`research/history/staging/measure-0830-rho-sup-z41.js`, a four-tail file\nembedded tail by tail with `qc/embed.js --streams both` (stdout is the\ndeterministic record; every timing figure is on stderr, by rule). No existing\nfile edited, no git command run. Every number below is in one of the four\nOUTPUT blocks or is quoted from an embedded artifact cited by file and\nsection, except the arithmetic named in §8. Calibration per claim: PROVEN,\nMEASURED, HEURISTIC, OPEN, REFUTED.)*\n\n## 0. What is open, first\n\n- **RML(α) is open at every α, REC(s, u₀) is open at every (s, u₀), and no\n exponent moves here.** One more exact point on a measured curve cannot\n decide a statement quantified over all z ≥ z₀. Whatever the eighth point\n does to the seven-point slope 2.7660 ± 0.2120\n (`attack-0829n-rml-proof.md` OUTPUT S5), it moves a fit. A point that\n lands the slope at or above 3.0 within one standard error refutes nothing\n asymptotic, and a point that lands it below 3.0 by two standard errors\n proves nothing. That was written before the run (§5).\n- **The only kill on the table is finite-z**: REC(3.0, 3.0) is false AT\n z = 41 iff the walked sup|R_H| at H = 41³ reaches HM − 1\n (`attack-0829n-rml-proof.md` §6, first bullet). Its forecast margin is a\n factor above ten (§5), and the expected outcome, stated in advance, was\n \"not killed\".\n- **The exact z = 41 point does NOT exist after this note.** The run that\n answers the question priced at 4.8 to 5.4 h of wall clock on this machine\n from measured per-position costs (§4), above the 3.5 h rule, so it was\n not run; what exists is the engine, its custody at every lower level, the\n pre-registration, one pilot segment of twelve (tail 2, a lower bound), and\n the segmented plan for the compute box. The eleven remaining segments are\n an ask.\n- **The sup at z = 41, when it exists, will be one number from one engine.** The engine is new in\n three places (§2), each proven to compute the same object, and each checked\n against the old engine at every level where both can run (§3). There is no\n independent second implementation at z = 41 itself; the O(n) recomputation\n at the argmax is a self-check, not an independent one.\n- **z = 43 and 47 stay out of reach for the exact sup** by the same\n accounting as before (`rho-exact-z31-01.md` §6): 41× and 1763× the z = 41\n work respectively, and nothing in §2 changes the linear-in-W law.\n- **The brief carried one error, recorded in §7 (checks): the seven exact\n points do not span \"less than an octave\" of z.** log₂(37/13) = 1.509, so\n they span one and a half octaves; the phrase is the corpus's own\n (`attack-0829n-rml-proof.md` §0, §5; `rho-exact-z31-01.md` §0) and it is\n wrong there too. Adding z = 41 extends the span to log₂(41/13) = 1.657\n octaves, not to two.\n\n## 1. The object, unchanged\n\ns = 3.0, D = z^s, terms from `sift-limit-lemmaV.js`'s `buildTerms(z, D)`,\nthe identical call at every exact level; ρ(y) = Σ_j w_j ψ((y − c_j)/q_j),\nψ(t) = t − ⌊t⌋ − 1/2, period W = P(z) = ∏_{p seg-k.out 2> seg-k.err` (segment 12 is done); each segment is ~3.09e11\npositions and independent; the table (6.47 GB shared) is rebuilt per\nsegment. On this machine a segment is ≈ 3.09e11 × 4.89 ns + 453 s ≈ 33 min;\non the box, per-core Zen 3 against M1 P-core is unmeasured (run `bench/`\nfirst per the box protocol), and with 16 cores at 0.7× per core the eleven\nsegments price at roughly 11 × 33 × (10/16)/0.7 ≈ 5.4 h, overnight class,\nnice-able, 7 GB resident. Ferry the 22 output files back; bind each as a\ntail via the declared-provenance path with the box's node version and\nhardware recorded; combine by max (sup|ρ̃|, sup|R_H3|, sup|R_H4|) and by sum\n(Σ R_H², positions) over the twelve tails, and check Σ positions =\n3,710,374,718,933. Nothing in this plan changes the object.\n\n## 5. Pre-registration, written to tail 1 before any z = 41 segment ran\n\nTail 1, S4, embedded (out-sha256 78984bb6…) before tail 2 was launched.\n\n- **(i) The seven-point slope** recomputed on this file's own S1 column:\n 2.7660 ± 0.2120, matching `attack-0829n-rml-proof.js` S5. Span: 1.509\n octaves of z, not \"less than one\".\n- **(ii) OLS forecast** of sup|ρ̃|(41): 60.07; 1-se prediction band\n [47.25, 76.37]; 2-se band [37.16, 97.11].\n- **(iii) Gaussian-law forecast:** C_true(41) inside the seven-point range\n [0.6044, 0.8412], i.e. sup|ρ̃|(41) in [55.83, 77.69]; F4 (C_true ≤ 1) is\n false at z = 41 iff sup|ρ̃|(41) > 92.36. sup/rms forecast for ρ̃:\n [4.653, 6.476]; for R_H3, from the five-point trend (vi): 5.646, band\n [4.869, 6.546].\n- **(iv) Kill rule for the u₀ = 3 form, finite-z:** REC(3.0, 3.0) is FALSE\n AT z = 41 iff sup|R_H3|(41) ≥ HM − 1 = 1807.417. Forecast: not killed,\n with F3 ≥ 9.31 even at the 2-se top of the band; the kill needs\n sup|ρ̃|(41) ≥ 903.71, fifteen times the forecast.\n- **(v) What the eighth point does to the fit:** on the forecast, the\n eight-point slope is 2.7660 ± 0.1656; at the 1-se band ends 2.6562 /\n 2.8757; at the 2-se ends 2.5465 / 2.9855. The eight-point fit excludes\n u₀ = 3.0 from below (slope − se ≥ 3.0) iff sup|ρ̃|(41) ≥ 343.0, and from\n above (slope + se ≤ 3.0) iff sup|ρ̃|(41) ≤ 68.9. Forecast: neither. Either\n way it moves a fit and refutes nothing asymptotic.\n- **(vi) sup|R_H3|** five-point slope 2.4522 ± 0.2095; forecast at 41:\n 58.26, band [50.24, 67.56].\n- **(vii)** The allowance REC(3,3) grants at z = 41 is 41^{1.5} = 262.53\n against a forecast ratio of 5.6; no ε is in force, so only (iv) is a kill.\n\n**Scoring rule, fixed here:** the pilot segment (tail 2) covers a twelfth of\nthe half range and so yields LOWER BOUNDS only; none of (ii)–(vi) can be\nscored against it, and the note will not score them. They are scored when\nthe twelve tails exist.\n\n## 6. The pilot segment at z = 41 (tail 2): lower bounds, the fleet rate, and nothing scored\n\nTail 2, S5: segment 12 of 12, 101 of the 1201 chunks (every twelfth plus\nthe wrap chunk), 309,200,483,133 positions, the count matching the plan;\nworst block drift 6.0e-9; the sup's argmax recomputed exactly in O(n) to\n5.7e-9. What it measured, every figure a bound on the full-period value\nand none of them the value:\n\n- **sup|ρ̃|(41) ≥ 70.651250**, attained at position 1,704,538,822,781. That\n lower bound already sits inside the pre-registered 1-se band\n [47.25, 76.37] and above 68.9, the level below which the eight-point fit\n would exclude u₀ = 3.0 from above (§5 (v)); so that exclusion is already\n impossible, and no other forecast is scored here (§5's scoring rule).\n- **sup|R_H|(41) ≥ 64.416936 at H = z³ and ≥ 97.905636 at H = z⁴.** The\n segment's F3 = 28.058 is an UPPER bound on the full-period margin; the\n kill (sup|R_H3| ≥ 1807.417) is not reached in this twelfth and cannot be\n ruled out from it.\n- The segment's m2 = 144.009169 against the full-period closed form\n 143.927339 (S2), 0.06% apart on one twelfth of the range [ARITHMETIC];\n the segment mean 8.96e-4 against the proven 0.\n- **The fleet rate at z = 41 itself** (stderr): 3.092e11 positions in\n 1398.6 s = 4.52 ns/position of wall with 10 workers, and the table build\n 164.6 s. [ARITHMETIC] The remaining eleven segments price at 11 ×\n (1398.6 + 164.6) s = 4.8 h on this machine, which confirms the §4\n decision from the z = 41 rate rather than the z = 37 one; one segment is\n 26 min.\n\nRung: MEASURED, one twelfth of the half range, lower bounds only.\n\n## 7. Brief-claim checks\n\n| brief claim | record | verdict |\n|---|---|---|\n| period P(41) = 7.4e12 | W = 7,420,738,134,810 (S3; `scanstat-t37.md` header) | holds |\n| existing exact points z = 13..37, seven | `rho-exact-z31-01.js` S2, seven rows | holds |\n| slope 2.766 ± 0.212, 1.1 se under u₀ = 3, 5.8 se under u₀ = 4 | `attack-0829n-rml-proof.js` S5: 2.7660 ± 0.2120, 1.10 se, 5.82 se | holds |\n| \"seven points spanning less than an octave\" | log₂(37/13) = 1.509 | WRONG, in the brief and in the corpus (§0) |\n| z = 31, 37 walks cost 6928 s | `rho-exact-z31-01.md` §4: 6928.4 s for the whole producer, Λ included | holds as stated there |\n| `scanstat-t37.md` found G₂(37#) over 7.4e12 positions \"in minutes\" | 5 shards × 35.2 min in lockstep over 2.18e11 slots | 35 min wall, a different object (gaps between survivors) |\n| compute box 5950X, 16 cores, 128 GB | memory file; not verified here | not verified |\n\n## 8. Readings\n\n1. **MEASURED, custody.** The new engine (reflection, P(29) table, per-block\n re-seeding) reproduces the cited exact sup|ρ̃| at all seven levels\n z = 13..37 to the cited six decimals in legacy mode and to 1.2e-7 in\n half-range reseed mode, and the three engines agree at z = 41 on three\n sample chunks to 1.4e-7. The reflection identity is PROVEN (§2(a)) and\n checked at every position of z = 13, 17, 19. Tail 1, S0, S2.\n2. **MEASURED, new columns at the old levels.** sup|R_H| at H = z³ walked\n for the first time: 8.600331, 15.737099, 21.611574, 29.537383, 47.838730\n at z = 19..37, between 0.92 and 1.30 times sup|ρ̃|, so the factor 2 that\n `attack-0829n-rml-proof.md` §5 stood in with was 1.5 to 2.2× loose; the\n certified margin at u₀ = 3 is F3 = 26.4 to 35.9 over five levels and is\n not falling. sup|R_H3|/rms runs 3.39 → 5.68 (log_z 0.41 → 0.48). Tail 1, S1.\n3. **MEASURED, the price.** Naive 69.8, table-legacy 23.5, table-reseed\n 36.6 ns/position per worker at z = 41; fleet 4.89 ns/position of wall at\n z = 37 in the mode the falsifier needs; the half range plus wrap is\n 3,710,374,718,933 positions, so the exact point is ≈ 5.4 h here, over the\n rule, and was not run. The obtained speed-ups are 2× (PROVEN reflection)\n and 2.9× (table); no bound-and-prune exists on the routes tried (§2(d)).\n Tail 1 stderr, S2, S3.\n4. **PRE-REGISTERED, unscored.** Forecast sup|ρ̃|(41) = 60.07, 1-se band\n [47.25, 76.37]; kill of REC(3,3) at z = 41 needs sup|R_H3| ≥ 1807.417,\n fifteen times the forecast; the eight-point fit cannot exclude u₀ = 3\n from below unless sup|ρ̃|(41) ≥ 343.0. Scored only when all twelve\n segments exist. Tail 1, S4.\n5. **MEASURED, lower bounds from the pilot twelfth.** sup|ρ̃|(41) ≥\n 70.651250, sup|R_H3|(41) ≥ 64.416936, sup|R_H4|(41) ≥ 97.905636; the\n fleet rate at z = 41 is 4.52 ns/position of wall, so the eleven unrun\n segments are 4.8 h here or an overnight run on the box. The one\n pre-registered outcome already decided is negative: the eight-point fit\n cannot exclude u₀ = 3 from above (needs sup ≤ 68.9). Tail 2, S5.\n6. **Corrected, not applied.** \"Seven points spanning less than an octave\"\n (brief; `attack-0829n-rml-proof.md` §0, §5; `rho-exact-z31-01.md` §0) is\n arithmetically wrong: log₂(37/13) = 1.509 octaves. Recorded here for the\n adjudicator; no live file edited.\n\n## 9. Not reached\n\n- **The exact sup|ρ̃|(41), sup|R_H3|(41), sup|R_H4|(41):** eleven of twelve\n segments unrun; plan in §4. Until then nothing in §5 is scored.\n- **z = 43, 47:** 41× and 1763× the z = 41 work; a different object is\n needed, not a faster walk (`rho-exact-z31-01.md` §6, unchanged).\n- **A bound-and-prune certificate:** not found; §2(d) is a negative on\n three routes, not a theorem.\n- **sup|R_H| at u₀ = 3.5 and 4.25:** dropped from the trackers to buy\n speed; only u₀ = 3 and 4 are walked.\n- **An independent second implementation at z = 41:** none; the argmax\n recomputation is a self-check.\n\n## 10. Custody and gate\n\nProducer `research/history/staging/measure-0830-rho-sup-z41.js`, four\ndeclared tails. Tail 1 (stage A) bound by `node research/qc/embed.js\n--streams both --tail 1 … -- --stage A`, code-sha256 61af2c79…, out-sha256\n78984bb6…, 201 body lines, elapsed 2554.6 s. Tail 2 (stage B, segment\n12/12) bound the same way; its fingerprint is in the file. Tails 3 and 4\nare reserved and unbound (the segments belong on the box). A `--check`\nre-run of tail 1 costs 43 minutes and was not bought. Derived arithmetic in\nthis note, done on embedded columns and not by the script: the ratios and\n\"×\" factors in §3, every wall-clock extrapolation in §4, and the octave\ncounts. `node research/qc.js --full` was run at the end; its result is in\nthe final report. The ledger gate requires TODO.md item 0 to list\n`Q-rho-sup-z41-0830`; this note may not edit TODO.md, so that line is the\norchestrator's.\n", "research/history/staging/phase1-T3prep-decision-rule.md": "# T3-prep: the pre-registered decision rule for \"does G2(x#)/x² fall?\"\n\n\n\n*(2026-08-18, phase 1. Read-only on the corpus; no repo file edited. Scratch\ncode lives outside the repo. Every number below was recomputed here from\nscratch and the calibration reproduces attack 1 digit for digit.)*\n\n**HEADLINE.** The instrument is calibrated and sharp (it resolves the one-class\ncontrol at 11.6 σ on the same nine points where it returns 0.9 σ on G2), the\nhonest band on the slope is **±0.117 at nine terms**, and the decision rule\nbelow is therefore pre-registered; but the power analysis says the two live\nhypotheses separate at **x = 53** only if the falling law is x·ln²x, and at\n**x = 151** if it is the repo's own conjectural x·ln³x, so **phase 1's extra\nterms at 43 and 47 cannot settle anything, and no reachable exact ladder can\nsettle the version of the question the repo actually believes.**\n\nThe reason is one number: on x ∈ [11, 41] the noiseless log-log reading of\n**c·x·ln³x is −0.009** where flat reads 0.000. The conjectured truth and the\nthing it is being tested against are the same measurement on this range. That\nis not a limitation of the estimator. It is a property of the interval.\n\n---\n\n## 1. THE RULE. Apply this mechanically; nothing below is needed to use it.\n\n### 1.1 Inputs\n\n- A set of **exact** G2(x#) values at primes x. Nothing else. Greedy Y2 values\n are excluded here and handled separately in §1.5.\n- The primary window is **x ∈ [11, X₁]**, all primes in it, no gaps, minimum\n **n = 9**. X₁ is the largest x with an exact term.\n- x_min = 11 is frozen. It is not a tuning knob. See §1.6 for why, and for the\n mandatory robustness row at x_min = 5.\n\n### 1.2 The estimator\n\nLet `b` = the OLS slope of `ln(G2(x#) / x²)` against `ln x` over the window,\nand `se_nom` its residual standard error, `se_nom = sqrt(RSS/(n−2)/Sxx)` with\n`Sxx = Σ(ln x − mean ln x)²`.\n\n**The denominator is x², and x is the same variable used as the regressor.**\nNot x′² (nextprime), not x² − x. Those are legal as *levels* and forbidden as\n*trends*: see §5.1, where the artifact is computed exactly and is 0.199.\n\nEquivalently `b = a − 2` with `a` the fitted exponent of G2. Report both.\n\n### 1.3 The band\n\n```\nse_honest = max( 1.55 * se_nom , 1.052 / n )\n```\n\nBoth terms are measured on the 58-term control, not assumed (§4). At n = 9 they\nagree: 0.116 and 0.117. Use the larger. **Never quote se_nom alone**; on the\ncontrol it covers the true error in only 8 of 14 nested designs, while\n1.55·se_nom covers 14 of 14.\n\nFor n > 40 the inflation factor 1.55 is too small. The control's own\nsliding-window table gives 2.0 at width 15, 2.2 at width 25 and 2.3 at width\n30. Re-measure it before applying the rule at n > 40; do not extrapolate.\n\n### 1.4 The verdict\n\n| condition | verdict |\n|---|---|\n| `b + 3·se_honest < 0` | **FALLS** |\n| `b − 3·se_honest > 0` | **RISES** |\n| otherwise | **FLAT / UNRESOLVED** |\n\nThree gates must all pass before FALLS or RISES may be reported:\n\n1. **n ≥ 9** and the window contains every prime in [11, X₁].\n2. **Sign agreement.** The same fit over x ∈ [5, X₁] must have the same sign as\n the primary fit. It need not be significant. If the signs differ the verdict\n is FLAT / UNRESOLVED whatever the primary says.\n3. **The verdict names its window.** \"G2/x² falls\" is not a sentence. \"b =\n −0.31 ± 0.06 over x ∈ [11, 97], n = 21\" is.\n\nThreshold justification: 3 σ with this band gives a measured false-positive\nrate of 0.000 under the flat law at every n ≤ 58 in a 20,000-replicate Monte\nCarlo with the control's own residual structure (§6). Yesterday's failure was a\nfalse positive. The rule is deliberately asymmetric against that failure.\n\n### 1.5 Greedy / certificate values: the one inference they license\n\nY2 is a **lower** bound: `G2(x#) ≥ Y2(x) + 1`.\n\n- **Y2 values may never appear in the same fit as exact values.** Pooling the\n four published Y2 levels onto the nine exact terms moves the slope from\n −0.069 to −0.498, which is **5.7 times the object's own standard error**, and\n turns FLAT into a 14.7 σ \"FALLS\". That is a manufactured verdict.\n- A Y2 ladder fitted alone may be reported, with this sentence attached at the\n point of use: *a falling lower bound is consistent with the truth falling,\n being flat, or rising, and therefore carries no verdict.*\n- **The one exception, and it is one-directional.** If a certified Y2 ladder\n ever satisfies `Y2/x² − 3·se_honest > 0` with Y2/x² growing without bound,\n the truth rises too, and the Zone Postulate is refuted. **A lower-bound\n ladder can produce a RISES verdict and can never produce a FALLS verdict.**\n\n### 1.6 Forbidden operations\n\n- **No slot-unit ratio across natals.** At fixed b = 37 the quantity L·m̄/G2\n runs 0.985, 0.907, 0.804, 0.695, 0.631, 0.531 across natals 5, 11, 13, 17,\n 19, 23 while the object is byte-for-byte identical. A ladder that changes\n natal as it climbs manufactures a 46% fall out of tile geometry. Every fit\n must be at one natal, and the natal must be stated. Working in G2/x² directly\n avoids this entirely, which is the main reason the rule is written in G2 and\n not in combined L.\n- **No control-corrected reading as a verdict.** Subtracting the control's bias\n assumes G2 has the control's shape, which is the question. Report it as a\n sensitivity, never as the verdict.\n- **No bisection on feasibility unless monotonicity is proved.** For the exact\n covering formulation it *is* provable and here is the proof: if a choice\n (a_p) covers [0, L) then the same choice covers [0, L−1), a subset, so\n feasibility is monotone decreasing in L. A greedy predicate has no such\n proof, which is exactly why `two-class-lower-bounds.js`'s `maxM` is unsafe.\n- **No verdict from a comparison of means.** The rule takes a slope and a band.\n\n### 1.7 What the rule returns today\n\nn = 9, x ∈ [11, 41]: b = **−0.0694**, se_nom 0.0747, se_honest **0.1169**,\n|b|/se_honest = **0.59 σ**. Sign gate: the [5, 41] fit is −0.2025, same sign,\npasses. **Verdict: FLAT / UNRESOLVED.**\n\n---\n\n## 2. The question, restated so it is answerable\n\nThe criterion \"combined L against the window in slot units\" was shown by attack\n1 §5 to be G2(x#)/x² at natal 5 to within 1.1% by b = 41, and §7 showed that\ncriterion is TPC-equivalent. So \"does it fall below the threshold\" is not\navailable. The answerable question is:\n\n> **Over the exactly-known range, what is the local logarithmic slope of\n> G2(x#)/x², and is it distinguishable from zero at the measured band?**\n\nThat is an estimate, not a limit. What each answer licenses:\n\n| verdict | licensed |\n|---|---|\n| FALLS | \"G2/x² is measurably decreasing over [11, X₁].\" Nothing about the limit. |\n| RISES | The same, upward. If sustained and unbounded this refutes the Zone Postulate, which is the one direction where finite data can be decisive. |\n| FLAT | \"Over [11, X₁] the object is within the band of x². The asymptotic exponent is untouched.\" |\n\n**No verdict of this rule is evidence about the asymptotic exponent.** The\ncontrol proves the point at full strength: 58 terms of an object whose exponent\nis 1 return 1.282 with white residuals and no drift\n(`research/exponent-control.md` §1). A clean fit at these sizes carries no\nasymptotic information, and no reachable ladder length changes that.\n\n### 2.1 The better parametrisation, and the number to quote\n\nWrite the truth as `G2 = c·x·ln^k x`. On design points with\n`s = OLS slope of ln ln x on ln x`, the noiseless reading is `b = k·s − 1`. So\n\n```\nk_hat = (b + 1)/s se_k = se_honest/s k_flat = 1/s\n```\n\n`k_flat` is the log-power at which the law is numerically **identical to x²**\non that range. It is the resolving power of the range, in one number.\n\n| range | n | s | k_flat | k_hat measured |\n|---|---|---|---|---|\n| [11, 41] | 9 | 0.3302 | **3.028** | **2.82 ± 0.35** |\n| [5, 41] | 11 | 0.3863 | 2.589 | 2.06 ± 0.26 |\n\nRead the top row. **The repo's conjectural truth k = 3 sits 0.51 σ from the\nmeasurement, and \"flat\" sits 0.59 σ from it, on the same side.** The two are\nnot rival readings of this data. They are the same reading.\n\nThe bottom row is why x_min is frozen: moving it from 11 to 5 moves k_hat by\n0.76 and moves k = 3 from 0.5 σ to 3.65 σ. The range choice is worth more than\nthe answer, so it must be chosen before the data and never afterwards.\n\n---\n\n## 3. Calibration: the instrument reproduces, term for term\n\nWritten from scratch, no repo code reused.\n\n| quantity, x ∈ [11, 41], n = 9 | here | attack 1 |\n|---|---|---|\n| control h/x², slope | **−0.8209 ± 0.0709** | −0.821 ± 0.071 |\n| object G2/x², slope | **−0.0694 ± 0.0747** | −0.069 ± 0.075 |\n| control detection | **11.6 σ** | 11.6 σ |\n| object detection | **0.9 σ** | 0.9 σ |\n| separation | **7.3 σ** | 7.3 σ |\n| d ln G2/d ln x | **1.9306 ± 0.0747** | 1.931 ± 0.075 |\n| control h/x² on [5, 41], n = 11 | **−0.8163 ± 0.0397** | −0.816 ± 0.040 |\n\n**CALIBRATION VERDICT: PASS.** The instrument resolves a genuine fall at 11.6 σ\non nine points. The absence of signal on G2 is a measurement, not a lack of\npower, at any effect size down to about 0.35 in slope units.\n\n### 3.1 Data custody, three independent recomputations\n\n- **The control h(x#) = A048670, eleven terms recomputed from scratch** by\n direct segmented sieve over the whole period, x = 2 to 31 (31# = 2.01·10¹¹).\n Survivor count matched the closed form φ(x#) exactly at every level, and all\n eleven maxima match the repo array `research/exponent-control.js:const H`.\n h(37#) = 66 and h(41#) = 74 are the repo/OEIS values only and are flagged as\n such; they are two of the eleven points in the [11, 41] control fit.\n- **G2 itself recomputed by exact covering search**, ten terms, x = 2 to 29:\n 2, 6, 12, 30, 42, 66, 108, 150, 204, 258. Every one matches. This is a third\n independent route to the ladder (enumeration, bit-parallel, covering) and it\n is also a direct check of the brief's CRT identity `G2(x#) − 1 = the maximum\n coverable interval`, which the search uses as its definition.\n- The sliding-window band constant reproduces the repo's stated 1.05/width at\n **1.052** from a twelve-width least-squares fit.\n\n---\n\n## 4. The band, determined from the data\n\nFour independent routes, all measured.\n\n**(a) Point scatter of the ratio, x ∈ [11, 41].** G2/x²: mean 0.3658, sd\n0.0344, **cv 9.4%**, range 0.3068 to 0.4155. This is the level band, not the\nslope band, and it is not what a trend must beat.\n\n**(b) Residual scatter about the log-log line**, which is what the slope sees:\nobject rms(ln) 0.0858, control 0.0814 on the same nine x. The two objects have\nthe same scatter where both are known, which is what licenses using the\ncontrol's long baseline as G2's noise model.\n\n**(c) Sliding-window scatter on the 58-term control**, the honest\nbetween-window band:\n\n| width | windows | sd(win) | mean se_nom | VIF |\n|---|---|---|---|---|\n| 7 | 50 | 0.1644 | 0.1064 | 1.54 |\n| 9 | 48 | 0.1091 | 0.0771 | 1.41 |\n| 11 | 46 | 0.0884 | 0.0585 | 1.51 |\n| 13 | 44 | 0.0827 | 0.0458 | 1.81 |\n| 19 | 38 | 0.0575 | 0.0275 | 2.09 |\n| 30 | 27 | 0.0366 | 0.0158 | 2.32 |\n\nFit: `sd(win) = 1.052/width`. Detrending against the control's own smooth shape\nbarely changes it (K goes 1.05 to 1.10), so **the between-window scatter is\nnoise, not the drift of the local exponent.** The nominal se understates by\n1.4 to 1.6 at the widths that matter and by more than 2 at width 30.\n\n**(d) Jackknife on the object**, n = 9: se 0.0885 against nominal 0.0747, ratio\n1.19. The most influential term is x = 11; dropping it moves the slope to\n−0.139.\n\n**(e) How the noise scales with x, measured.** Leave-one-out local residuals of\nthe control, in four blocks: rms 0.098 at x ≈ 20, 0.049 at 78, 0.024 at 144,\n0.018 at 226. Power fit **sd(x) = 0.894·x^−0.711**. Extreme-value theory\npredicts 1.2825/θ(x), which decays faster, so the fitted law is the\nconservative one. The practical consequence is large and favourable: the\nper-point noise at x = 113 is a third of what it is at x = 41.\n\n### 4.1 How many terms for a fall of a given size\n\n`n ≥ 3K/effect` with K = 1.052, where \"effect\" is the noiseless slope the law\nwould read on that range.\n\n| effect \\|b\\| | n for 3 σ | n for 5 σ | a law with that effect |\n|---|---|---|---|\n| 0.60 | 6 | 9 | k = 1 on [11, 41] |\n| 0.34 | 10 | 16 | k = 2 on [11, 41] |\n| 0.20 | 16 | 27 | k = 3 on [11, 180] |\n| 0.16 | 20 | 33 | k = 3 on [11, 113] |\n| 0.05 | 64 | 106 | k = 3 on [11, 55] |\n| 0.01 | 316 | 526 | **k = 3 on [11, 41], i.e. today** |\n\nThe last row is the whole problem stated as an arithmetic fact. Detecting the\nrepo's conjectured law against flat on today's range would need 316 exact\nterms. There are thirteen.\n\n---\n\n## 5. Trap defences, with the artifacts computed exactly\n\n### 5.1 Trap: the denominator convention. Artifact = 0.199, exactly.\n\nSame nine terms, three denominators:\n\n| quantity | slope | σ from 0 |\n|---|---|---|\n| G2/x² | **−0.0694** | 0.93 |\n| G2/(x² − x) | −0.1216 | 1.66 |\n| G2/x′² (nextprime) | **+0.1297** | 1.75 |\n\n**The sign flips.** The cause is exact and needs no statistics: the OLS slope of\nln x′ on ln x over these nine points is **0.9004**, not 1, so using x′² as the\ndenominator while regressing on ln x subtracts 2 × 0.9004 instead of 2. The\nartifact is `2(1 − 0.9004) = 0.199`, and −0.0694 + 0.199 = +0.1296 against the\nmeasured +0.1297, agreeing to one part in 10⁴. The artifact is **1.7 times the\nhonest band**. This is the same index-convention effect `exponent-control.md` §6\nrecords as flipping the h2 margin trend from −0.084 to +0.154.\n\nRule: the denominator's variable must be the regressor. x′² is the right\n*threshold* for the Zone Postulate and the wrong *denominator* for a trend.\n\n### 5.2 Trap: a falling lower bound. Artifact = 5.7 standard errors.\n\nY2/x² over the four published levels x = 37, 229, 1009, 4001 falls at\n−0.5242 ± 0.0458, an 11.5 σ detection. Pooling those onto the nine exact terms\ngives −0.4978 ± 0.0340, 14.7 σ, \"FALLS\". The pool moves the answer by 0.4284,\n**5.7 times the object's own nominal se**. §1.5 forbids it.\n\n### 5.3 Trap: slot units across natals. Artifact = 46%.\n\nReproduced from attack 1 §5 at fixed b = 37: L·m̄/G2(37#) = 0.985, 0.907,\n0.804, 0.695, 0.631, 0.531 at natals 5, 11, 13, 17, 19, 23. The object does not\nchange. The rule works in G2/x², which has no natal at all.\n\n---\n\n## 6. Power: where the two hypotheses separate\n\nMonte Carlo, 20,000 replicates per cell, noise = the measured heteroscedastic\nlaw `sd(x) = 0.894 x^−0.711` applied to a stationary block bootstrap of the\ncontrol's own standardised residuals (mean block 5, lag-1 autocorrelation 0.47\nin the standardised residuals, so the correlation structure is real and\ncarried). Rule as pre-registered in §1: BAND-A, 3 σ.\n\n**P(rule says FALLS), by truth and by largest exact term X₁:**\n\n| X₁ | n | flat | k=2 | k=2.5 | k=2.8 | k=3 | k=3.5 |\n|---|---|---|---|---|---|---|---|\n| **41 (today)** | 9 | 0.000 | 0.442 | 0.068 | 0.003 | **0.000** | 0.000 |\n| **43** | 10 | 0.000 | 0.585 | 0.090 | 0.011 | 0.000 | 0.000 |\n| **47 (end of phase 1)** | 11 | 0.000 | 0.699 | 0.137 | 0.019 | **0.000** | 0.000 |\n| 53 | 12 | 0.000 | **0.823** | 0.220 | 0.050 | 0.004 | 0.000 |\n| 71 | 16 | 0.000 | 1.000 | 0.690 | 0.235 | 0.072 | 0.000 |\n| 89 | 20 | 0.000 | 1.000 | 0.957 | 0.651 | 0.243 | 0.000 |\n| 113 | 26 | 0.000 | 1.000 | 1.000 | **0.980** | 0.742 | 0.011 |\n| 151 | 32 | 0.002 | 1.000 | 1.000 | 1.000 | **0.987** | 0.138 |\n| 251 | 50 | 0.000 | 1.000 | 1.000 | 1.000 | 1.000 | **0.952** |\n\n**First X₁ reaching 80% power: k = 2 → 53. k = 2.5 → 79. k = 2.8 → 113.\nk = 3 → 151. k = 3.5 → 251.**\n\nFalse positive under flat: at most **0.003** at every n ≤ 58, and exactly 0.000\nat every n ≤ 21, which is the regime phase 1 lives in. At n = 165 it reads 0.025,\nbut that row recycles a 56-term residual pool through the bootstrap and its\ncorrelation structure is not trustworthy; treat n > 56 as unvalidated.\n\n### 6.1 The forecast for phase 1's own terms, and it is negative\n\nAppending noiseless model values at 43 and 47 to the nine real terms, rescaled\nthrough the real x = 41 point, the rule returns:\n\n| truth | b at n = 11 | se_honest | verdict |\n|---|---|---|---|\n| c x² | −0.0997 | 0.0872 | FLAT / UNRESOLVED |\n| c x ln²x | −0.1202 | 0.0925 | FLAT / UNRESOLVED |\n| c x ln³x | −0.1086 | 0.0891 | FLAT / UNRESOLVED |\n| c x^1.8 | −0.1084 | 0.0891 | FLAT / UNRESOLVED |\n\nThe five hypotheses tested span **0.033 in the fitted slope**, against an\nhonest band of 0.089. **The two new terms move the answer by a third of the\nnoise, and they move it by the same amount whatever the truth is.** G2(43#) and\nG2(47#) are worth having for the ladder and for the OEIS submission. They\ncannot move this question, and the rule above will say so.\n\nThis confirms `exponent-control.md` §8's measurement from the other direction:\nadding an eleventh term to a ten-term control fit moves the exponent by 0.022\non average and 0.078 at worst.\n\n### 6.2 The largest k a range can call falling\n\n| X₁ | n | s | k_flat | largest k callable at 3 σ |\n|---|---|---|---|---|\n| 41 | 9 | 0.330 | 3.028 | **1.97** |\n| 47 | 11 | 0.322 | 3.106 | 2.22 |\n| 53 | 12 | 0.317 | 3.151 | 2.32 |\n| 71 | 16 | 0.302 | 3.307 | 2.63 |\n| 113 | 26 | 0.279 | 3.578 | 3.11 |\n| 151 | 32 | 0.269 | 3.721 | 3.32 |\n| 1009 | 165 | 0.202 | 4.943 | 4.80 |\n\n**x = 113 is the first range that can call k = 3 falling at all**, and 151 is\nthe first that does it with 80% power.\n\n### 6.3 What it costs to get there. Measured, not assumed.\n\nTrap 5 says cost estimates here err cheap, so both routes were probed.\n\n**Enumeration.** Attack 1's measured throughput is 1.188·10¹³ natal-19 slots in\n151 s on ten cores, i.e. 7.9·10¹⁰ slots/s, and the slot count is 0.0390·x#. The\nperiod multiplies by x each term:\n\n| x | x# | projected wall, 10 cores |\n|---|---|---|\n| 43 | 1.31·10¹⁶ | 1.8 h |\n| 47 | 6.15·10¹⁷ | 3.5 days |\n| 53 | 3.26·10¹⁹ | 186 days |\n| 59 | 1.92·10²¹ | 30 years |\n\n**Covering search**, which does not scale with the period. Measured here, naive\nimplementation, single core, \"cover the first uncovered point\" branching with\nan exact per-prime capacity bound:\n\n| x | wall | nodes to prove the maximum | ratio to previous |\n|---|---|---|---|\n| 17 | 0.004 s | 137,008 | |\n| 19 | 0.060 s | 1,532,047 | 11.2 |\n| 23 | 0.882 s | 20,052,980 | 13.1 |\n| 29 | 18.8 s | 323,924,673 | 16.2 |\n| 31 | **> 390 s, killed unfinished** | — | **> 21** |\n\nPer-term cost ratio **15 to 21 in wall, 11 to 16 in nodes**, and the x = 31\nstep passed 21 without finishing and was killed, against\nenumeration's factor of x. Ziller and Morack's optimised adversarial search\nachieves 6.7 per term (`h2-scoping.md` §3, digitised from their figure). Our\nfixed pair {0, −2} has p choices per prime rather than ~p²/2, so it should be\nstrictly cheaper than theirs. `h2-scoping.md` §2's structural finding also\nreproduces here independently, and sharper: **almost all of the cost is proving\nthat no longer interval exists**, not finding the optimum. At x = 23 that is\n35,232 nodes to find against 20,052,980 to prove, i.e. 0.18% against 99.8%,\nwhere `h2-scoping.md` measured 2.5% against 97.5% on the h2 solver. The\nasymmetry deepens with x in my run, which is the direction that matters: a\nlower-bound ladder for G2 stays cheap long after the exact one stops.\n\nReading, stated as a bracket because the extrapolation is four terms long:\n**x = 47 is easy by enumeration, x = 53 is a serious but real commitment,\nx = 59 needs the covering route with ZM-quality code, and x = 113 is not\nreachable by anything** (113# ≈ 10⁴⁶; ZM's own curve prices their term 30 at\n5.6·10⁵ machine-years).\n\n**So the separation point for the repo's own conjecture lies past the end of\nevery route that exists.** That is the finding this task was built to produce.\n\n---\n\n## 7. Two numbers now in circulation that need correcting\n\n**7.1 \"+0.05 ± 0.11 over eleven exact terms\" is a nine-term figure.**\n`attack-block-00-ADJUDICATION.md`:153-154 and the phase-1 shared brief both\ncarry that label. Attack 1's own text is explicit: \"the fitted slope … over the\n**nine** exact levels b = 11 to 41\" (`attack-block-01-ladder.md`:22-23), and its\nsummary table row reads \"b >= 11, nine terms\". The distinction is not cosmetic.\nFor the closest computable quantity, G2/x², the nine-term fit is −0.069 ± 0.075\nand the eleven-term fit is **−0.203 ± 0.064**, which is 3.2 σ on the nominal se\nand 2.0 σ on the honest band. A reader who applies the \"eleven terms\" label\nliterally gets a different verdict from the one the sentence reports.\n\n**7.2 \"the ratio bounces between roughly 0.24 and 0.32\" is G2/x′², not G2/x².**\nOver the thirteen exact terms: G2/x′² runs 0.222 to 0.314, G2/x² runs 0.307 to\n0.667, and restricted to x ≥ 11 it runs 0.307 to 0.416. The quoted band belongs\nto the nextprime convention, which §5.1 shows must not be used for a trend.\nStating a band without its denominator is the same class of slip the campaign's\ntrap 2 records.\n\n---\n\n## COVERAGE\n\n**What I did not reach.**\n\n- **h(37#) and h(41#) were not recomputed.** Direct sieving stops being cheap\n above 31# (2·10¹¹ in 7 minutes; 37# is 7.4·10¹² and 41# is 3.0·10¹⁴). Those\n two points are two of the eleven in the control fit that produces the 11.6 σ\n calibration, and they come from the repo array and OEIS. Nine of eleven are\n mine. If A048670 is wrong at n = 12 or 13 the calibration figure moves, and I\n did not check it independently.\n- **The covering search was not pushed past x = 29 to completion.** The 31 run\n was killed after 390 s without finishing. The per-term ratio is therefore\n measured over four steps and extrapolated over four more. Four-term\n extrapolations of cost curves are exactly what trap 5 is about, and mine is\n no better protected than the ones it corrects. The ratio is *rising*\n (11.2, 13.1, 16.2), which makes my x = 53 and x = 59 figures optimistic.\n- **The 1.55 inflation factor is not validated above n = 40.** I say so in the\n rule, but a rule with an unvalidated regime is a rule with a hole in it, and\n the hole is exactly where a future x = 100 campaign would live.\n- **No G2 term was computed above 41.** The forecast in §6.1 uses noiseless\n continuations, so it shows where the rule's centre lands, not the spread.\n\n**What I suspect but could not prove.**\n\n- The heteroscedastic noise law sd ∝ x^−0.711 is fitted on four blocks of the\n control and is well below the extreme-value prediction 1.2825/θ(x) ∝ 1/x.\n I think the truth is closer to 1/θ and that my power figures are therefore\n pessimistic, perhaps by one or two primes in X₁. I could not separate the\n two laws on 56 points.\n- The covering route probably reaches further than enumeration from x ≈ 47\n onward, but my implementation is naive by a factor I did not measure. A\n faithful reimplementation of the ZM branching is the single highest-value\n engineering item I can see for extending the exact ladder, and **no\n optimality-proving covering search for G2 exists in the corpus to price it\n against.** That absence is calibrated: `research/` was listed (132 scripts),\n the known positive `h2-scoping.md` hits on the ZM cost-curve query, and the\n two candidates are both explicitly lower-bound machinery\n (`two-class-lower-bounds.js` is the greedy; `h2-lower-ladder.js` states in its\n own header that it drops the optimality proof, which is 97.5% of the run, and\n therefore returns a certified lower bound). The cost curve in §6.3 is the\n first one measured for our object, and it is four steps long.\n\n**Where I am most likely wrong.**\n\n1. **The choice of x_min = 11 is the weakest load-bearing decision in the\n rule.** It moves k_hat by 0.76 and it is defended on grounds (band tightness,\n number of folded primes, consistency with attack 1) that are reasonable and\n not decisive. Someone could pre-register x_min = 5 with equal honesty and get\n \"falling at 2.0 σ\" instead of \"flat at 0.6 σ\". I have frozen it and stated\n the sensitivity rather than resolved it, and a rule whose answer depends on a\n frozen arbitrary choice is only half a rule.\n2. **The k parametrisation assumes the truth is c·x·ln^k x.** If the truth has a\n different shape (a genuine power, a log-log correction, a turnover) then\n k_flat is not the right resolving-power statistic and §6.2 is decoration.\n The family is the repo's own conjecture, which is a reason to use it and not\n a reason to believe it.\n3. **The 3 σ threshold buys its zero false-positive rate with power.** Under\n k = 2 at X₁ = 47 the rule fires only 70% of the time; a real fall of that\n size present in the data would be reported as unresolved three times in ten.\n That is the deliberate trade after yesterday, but it is a trade and it should\n be re-argued, not inherited, if the programme's cost of a missed detection\n ever rises above its cost of a false one.\n\n---\n\n*Scratch code: `verify-control.c` (control sieve), `g2cover.c` (exact covering\nsearch), `instrument.js`, `noise.js`, `power.js`, `power2.js`. None is in the\nrepo. No repo file was edited by this task.*\n", "research/history/staging/recon-0904-sifting-limit-floor.md": "# The sifting limit floor at kappa = 2: what a lower bound on beta(kappa) would have to be, and why the corpus already owns a measured barrier inside the band\n\n\n\n> **RIDER 2026-09-04 (orchestrator, from `redteam-0904-sifting-limit.md`).**\n> Four of five claims survive with corrections; the headline does not. The\n> \"METHOD ARTEFACT at rung MEASURED\" verdict is REFUTED as stated on four\n> grounds, each sufficient: the level-D LP fixes one ω-profile out of a class\n> Ford defines by a one-sided inequality, and a legal profile inside the\n> classical budget at x = 13 reads 5.0113 calibrated, above β₂, firing this\n> note's own pre-registered refutation; the 1.72 calibration is fitted at one\n> point and misses the second proven anchor β(1/2) = 1 by 66 to 82 per cent;\n> the pooled 3.3152 averages four statistics of which only s* is the\n> definitional analogue, and that route reads 3.9487 at x = 43 and is still\n> climbing (+0.43 per ln x, reaching 4.26645 at x ≈ 104 on the fit); and\n> 3.3152 is 22.3 per cent below 4.26645, not 18. The calibrated sentence: no\n> barrier theorem, no exhibited axiom-only argument inside the band, the LP\n> number is a finite-level measurement for one profile. Claim a CONFIRMED with\n> a second source off the OCR channel (Ford 2023 p. 37); Halberstam is p. 117,\n> not 116. Claim b CONFIRMED and strengthened: Brady's class (his p. 10) is the\n> axiom-only interval class Face 4 needs, and 1.819592 rests on proved facts\n> alone; quote it, not 1.8394. Claim c CONFIRMED but demoted: β(2) ≥ 2 is two\n> lines from Ford's one-sided (Ω) plus β(1) ≥ 2, so the padding construction is\n> correct and unnecessary. Of the eight live proposals two survived verbatim,\n> five were reworded and one (the LP floor as a statement about the class) was\n> refuted; the applied versions are in the live layer dated 2026-09-04.\n\n*(2026-09-04. Recon note, HELD. No existing file edited, no git command run\nbeyond read-only status and log. Proposed live-layer changes are listed in the\nlast section and are NOT applied. Legend as in `research/covering-dive.md`:\n**[PROVEN]** published theorem with source; **[VERIFIED]** checked\ncomputationally here; **[MEASURED]** empirical, finite range; **[DERIVED]**\nthis note's own deduction from sourced facts, not reviewed; **[CONJ]**\npublished conjecture; **[ABSENT]** searched and found nothing, with the\nconvention named.)*\n\n---\n\n## 0. Verdict\n\n**The disconfirming half first, and it is most of the note.** Nothing here moves\nan exponent. The open band `(2, 4.26645]` is untouched: no proven statement,\nlocated or derived, reaches into it from below, and the strongest thing that\nsits inside it is still the corpus's own calibrated LP floor, which is a\nfinite-`x` measurement with a wide reading range and not a theorem. The one\nderivation in this note that reaches the value `2` reaches the band's endpoint\nand stops there, and it is almost certainly folklore rather than anything new.\n\n**The corpus's `[ABSENT]` on this face is wrong as written, and wrong in a way\n`SEARCH-CONVENTIONS.md` predicts.** `research/sift-limit-attack.md`:143 records\n`[ABSENT — in sieve-theory vocabulary]` for the whole object and\n`SEARCH-CONVENTIONS.md`:162 records \"None found\" for a `κ = 2` extremal example.\nThe second is right and is now calibrated by a better source than our own\nsearch. The first is not: **lower bounds on `β(κ)` for `κ > 1` are in print, and\nhave been since 1991.** They were invisible because of a convention inversion.\nSelberg's *Lectures on Sieves* works with the reciprocal `a_k = 1/β_κ`, so his\n§17 is titled **\"Some upper bounds for sifting limits for constant sifting\ndensity\"** and its content is a **lower** bound on `β_κ`. A search for \"lower\nbound for the sifting limit\" is therefore a guaranteed clean negative against a\nsection whose title says the opposite. That is exactly the failure mode\n`SEARCH-CONVENTIONS.md` exists to name, and §1's table carries no row for this\nobject.\n\n**What is actually proven, and the best floor on `β(2)`.**\n\n| statement | value at `κ = 2` | rung |\n|---|---|---|\n| Selberg, *Lectures on Sieves* §17, in the reciprocal convention | `β_2 > 1.213` | **[PROVEN]** in print; read only through Google Books OCR snippets, constant uncertain |\n| Brady 2017 Cor. 1, `β_κ ≥ (1+o(1))·2κ/e` | `1.472` asymptotically, nothing effective at `κ = 2` | **[PROVEN]** in a deposited thesis, read at the page |\n| Brady 2017 Cor. 3, effective, stated for `κ ≥ 3` | vacuous at `κ = 3` (`0.6196`), first exceeds 1 at `κ = 4` (`1.1157`) | **[PROVEN]**, read at the page |\n| **Brady's own reduction evaluated at `κ = 2`, which he does not do** | **`β_2 ≥ 3e^{-1/2} = 1.81959`** from his Theorem 22 at `R = 1`; **`5/e = 1.83940`** using his computed `v_3 = 2` | **[DERIVED]** here, one step from his printed proof, unreviewed |\n| monotonicity of `β` in `κ`, by padding a `κ = 1` extremal problem | **`β(2) ≥ β(1) = 2`** | **[DERIVED]** here, §2; hypotheses listed, almost certainly folklore |\n| Granville's exceptional character, interval form | `β_interval(2) ≥ 2` | **[CONDITIONAL]** on Siegel zeros, `attack-barrier-kappa2.md` |\n| the corpus's calibrated one-point LP floor | `3.3152` pooled, individual readings `2.669` to `3.949` | **[MEASURED]**, `lp-push-x43.md`, and independently re-run here |\n\nSo the best **unconditional** floor on `β(2)` is `2`, and it rests on a\nderivation in this note rather than on a citation; the best floor resting on a\ncitation alone is `1.8394`, which is below the band. Under a Siegel-zero\nhypothesis the floor is also `2`, and that hypothesis implies the target.\n\n**Is the cap at 4.26645 a barrier or a method artefact? Method artefact, at\nrung MEASURED, with the class's own floor unknown.** The level-`D` linear\nprogram is the exact information-theoretic limit of every argument that reads\nonly the divisor-class counts, which is exactly what \"axiom-only\" means, and\nthis note's independent re-implementation of it reproduces all twelve of the\ncorpus's published cells to `4.23e-5` and then puts the calibration through a\ntest it had never faced: the calibrated reading must not exceed the published\nDHR value at its own `κ`, because the DHR sieve is itself inside the class.\n**0 of 15 readings exceed it**, and the headroom ratio at `κ = 2` is `0.7442` to\n`0.8344`. So `4.26645` sits about 18 per cent above the one-point floor at the\nlevels reachable, and is not the limit of what the axioms permit. What the\nclass's own limit **is** remains open: measured at about `3.3` with readings\nfrom `2.669` to `3.949`, proven only to be at least `2`.\n\n**The band that matters, restated.** `(2, 4.26645]` is the open band for a\ntheorem. `(2, ~3.3)` is the band in which the measured one-point floor and the\ntruth cannot be told apart on present evidence, and inside which any axiom-only\nargument would have to beat the corpus's own LP measurement to exist at all.\n\n---\n\n## 1. Sub-question 1: lower bounds on `β(κ)` for `κ > 1`, in the owning convention\n\n**The convention, first, because it decides the answer.** Selberg's sifting\nlimit is written `a_k` and is the **reciprocal** of `β_κ`: `a_{1/2} = 1`,\n`a_1 = 1/2`. An upper bound on `a_k` is a lower bound on `β_κ`. His table of\ncontents, read at a Google Books snippet of *Collected Papers* II\n(vol. id `JswcQj2B1HIC`, ISBN 9783540506263), lists\n\n> \"14. The sifting limit for constant sifting density … 16. Two examples in\n> connection with the B2R method 17. Some upper bounds for sifting limits for\n> constant sifting density 18. A historical digression, the parity principle\n> and a further example\"\n\nso the lower-bound content of the *Lectures* lives in a section whose title\nreads as its opposite in our convention. §16 is the examples section and its\nown opening sentence, read at a snippet, scopes it: \"We shall in this Section\nconstruct some particular examples or models in connection with the sifting\nproblem for constant density **when k = 1/2 and when k = 1**.\" **[PROVEN scope,\nread at OCR snippet]**\n\n**The bound itself.** §17 p. 201, at an OCR-degraded snippet in which the\nfraction bars are lost, reads `a_k < e^{(h-1)/k}/h`, optimised at `h = k` for\ninteger `k` to `a_k < e^{1-1/k}/k`, that is `β_κ > κ·e^{1/κ-1}`, and a variant\n\"for all `k > 1`\". At `κ = 2` that is `β_2 > 2e^{-1/2} = 1.213`. **[PROVEN in\nprint; the constant is OCR-uncertain, the direction and the `k > 1` scope are\nnot, and the asymptotic size `κ/e` is corroborated independently below.]**\n\n**The corroboration, and the effective form, read at the page.** Brady's 2017\nStanford dissertation (*Sieves and iteration rules*, Stanford Digital\nRepository `druid:gk881hk9239`, PDF text layer, and an identical copy at\n`notzeb.com/phd-thesis.pdf`) says at printed p. 11, immediately after his\nTheorem 7:\n\n> \"From this, one can improve a lower bound Selberg gave for the sifting limit\n> β_κ by a factor of 2. **Corollary 1.** β_κ ≥ (1 + o(1)) 2κ/e.\"\n\nand at printed p. 51:\n\n> \"**Corollary 3.** If β_κ is the sifting limit of a sieve of dimension κ ≥ 3,\n> then β_κ > (2⌊κ − √κ⌋ + 1) / e^{1 + 1/√κ}.\"\n\nEvaluated: `0.6196` at `κ = 3` (vacuous), `1.1157` at `κ = 4`, `1.1761` at\n`κ = 5`, `1.7120` at `κ = 6`, `3.4859` at `κ = 10`. So a bound above 1 is in\nprint for every `κ ≥ 4`. **[PROVEN, read at the page; a deposited dissertation,\nnot a refereed journal paper, and the derivation was read but not re-derived\nindependently here.]**\n\n**The mechanism, which is the part that matters for this programme.** Neither\nbound is an exhibited arithmetic sequence. Both come from Selberg's **model\nproblem** (*Lectures* §13, described in Brady's §1.3), in which all sifting\nprimes are assumed the same size, so a sieve depends only on the number of\nprime factors and the whole question collapses to: for which `v` does there\nexist `λ_0 = 1, …, λ_R` with `θ(n) = Σ λ_i C(n,i) ≤ 0` at every positive\ninteger `n` and `Σ_n θ(n) v^n/n! > 0`? `v_R` is the largest such `v`. Brady's\nTheorem 22, `v_R ≤ d + 2√d + 1` for `R = 2d+1`, is an upper bound on `v_R`, and\nan upper bound on `v_R` is a lower bound on `β_κ`. **This is the LP/obstruction\nroute of §3 in a different notation, and it is why no `κ > 1` example exists\nwhile `κ > 1` lower bounds do: at `κ > 1` the field bounds the sieve, not the\nsequence.**\n\n**Brady's reduction, evaluated at `κ = 2`, which he does not do.** His proof of\nCorollary 3 runs: suppose `β_κ < 2d+3`; take `P` the primes between `y^{1/(2d+3)}`\nand `y^{1/β_κ}`; then `v = Σ_{p∈P} κ/p = (κ+o(1))log((2d+3)/β_κ)`; the product of\nany `2d+3` of them exceeds the level, so a nontrivial lower bound sieve forces\n`κ·log((2d+3)/β_κ) ≤ v_{2d+2} = v_{2d+1}`. For each `d` this gives\n`β_κ ≥ (2d+3)·exp(-v_{2d+1}/κ)` whether or not the supposition holds, since if\n`β_κ ≥ 2d+3` the conclusion is stronger still. At `κ = 2`:\n\n- `d = 0`, `v_1 ≤ 1` by his Theorem 22: `β_2 ≥ 3e^{-1/2} = 1.81959`.\n- `d = 1`, `v_3 = 2` from his computed `v_R` table, printed p. 46:\n `β_2 ≥ 5/e = 1.83940`.\n- `d = 2`, `v_5 = 3.11714` from the same table: `β_2 ≥ 1.4730`, worse.\n\nThe `d = 1` and `d = 2` values rest on a table Brady computes by his Algorithm 1\nrather than on a theorem, so only the `d = 0` value `1.81959` rests on a proved\ninequality (his Theorem 22 at `R = 1`).\n\n**[DERIVED]**, one substitution from a printed proof, not reviewed. Corollary 3\nis stated for `κ ≥ 3` only because its optimised `d = ⌊κ−√κ⌋−1` is negative at\n`κ = 2`; the underlying inequality has no such restriction. Neither value\nreaches 2, so **nothing in print, and nothing one step from print, enters the\nband or even reaches its lower endpoint.**\n\n**What is genuinely absent, now calibrated by a better source than our own\nsearch.** No extremal example at any `κ > 1`. The best evidence for that is not\nour negative but Halberstam's own, in his review of Greaves, *Bull. AMS* **40**\n(2003) p. 116, read at the PDF text layer:\n\n> \"Concrete examples like Selberg's when κ = 1 and Iwaniec's when κ = ½ are not\n> known and greatly to be desired.\"\n\n**[PROVEN as a statement of the field's own state as of 2003; nothing located\nafter 2003 changes it.]** The DHR book states the `Ω(n) mod 2` example only in\nits `κ = 1` chapter (p. 95, Google Books snippet), and its index lists \"sieving\nlimit\" at pp. 79, 135, 144, 226, 228 with no lower bound and no optimality claim\nabove `κ = 1`.\n\n**Two published statements that look like lower bounds and are not.**\nGrupp–Richert's `ν(κ) > 2κ + 1` for `κ ≥ 7/2` (zbMATH review) bounds the\n**Ankeny–Onishi sieve's own** limit, and the DHR book's `α_κ > β_κ > 2` bounds\n**its own** delay-differential parameters. Both say a particular method cannot\ndo better. Neither is a statement about the axiom class. The corpus already uses\n`α_2 ≥ β_2 + 1` correctly in that method-specific sense\n(`sift-limit-attack.md`, the fractional-retention entry) and must not promote it.\n\n**Ford's notes, re-read for this question.** `sieve2023.pdf` from\n`ford126.web.illinois.edu`, §3.1 p. 37, text layer and rendered page agreeing.\nHis Definition 2 gives the class-level `β(κ)` this note uses. His Table 1 is\ncaptioned **\"Known upper bounds on the sieving limit β(κ)\"**, and the only lower\nbound anywhere in the 129 pages is `β(1) ≥ 2`. He also prints \"Selberg [137,\neq. (14.40)] showed that β(κ) ⩽ 2κ + 0.4454 for sufficiently large κ\", a\nconstant the corpus does not carry: `sift-limit-attack.md` §2 quotes Franze's\n`2κ + 19/36 = 2κ + 0.5278` for the same Selberg result. Both are at source; they\nare not the same number and the corpus should carry both or neither.\n**Ford does not say that no lower bound is known. He is silent, and the corpus\nhas been reading that silence as an absence.**\n\n**One quotation correction owed to Franze.** `sift-limit-attack.md` §2 reports\nFranze as recording \"Selberg's asymptotic `β_κ ≲ 2κ + 19/36`, approached rapidly\nfrom below: two independent technologies plateau near 2κ\". Read at the arXiv\nv1 PDF text layer and confirmed in the TeX source (`siftinglimitspaper.tex`,\nthe only version), Franze's line 73 is `\\beta_\\kappa \\lesssim 2\\kappa + 19/36`\nand his \"approached rapidly from below\" refers to **his own Λ²Λ⁻ table**\napproaching that asymptotic, not to the truth. His line 272 writes\n`\\beta_{\\kappa}\\gtrsim2u+1=2\\kappa+\\frac{19}{36}`, which contradicts his line 73\nand his abstract and is an apparent typo for `\\lesssim`; **it must not be\nquoted as a lower bound.** Also at source, and not in the corpus: \"The\ncalculations in Chapter 17 of [4] show that for the DHR sieves, `β_κ ≲ 2.44κ`.\"\nThe DHR table itself confirms the growth: `β_κ − 2κ` runs `0.266, 0.641, 1.072,\n1.535, …, 6.453` at `κ = 2 … 15`, so the DHR values do **not** plateau near `2κ`\nand the corpus's \"two independent technologies plateau near 2κ\" holds at small\n`κ` only.\n\n---\n\n## 2. Sub-question 2: is `β(κ)` non-decreasing, and does that give `β(2) ≥ 2`?\n\n**Not stated in any source read for this note.** Ford, Franze, Blight, Brady,\nthe DHR book and Halberstam's review all define `β` and none of the six asserts\nmonotonicity in `κ`. **[ABSENT in the six sources read; a weaker claim than\n\"absent from the literature\", and it should be read as the reach of this\nsearch.]**\n\n**The argument, attempted here.** Take a dimension-`κ` sieve problem `A` with\nlevel `y`, sifting parameter `z`, `s = log y/log z`, and normalised sifted\ndensity `S(A,z)/(|A|V_κ(z)) → 0`. Take an auxiliary sequence `B` of dimension\n`κ_B` with level at least `y`, and form the product sequence\n`A' = {ab : a ∈ A, b ∈ B}` with multiplicity. Then:\n\n1. **Dimension adds.** For squarefree `d`, `p | ab` iff `p | a` or `p | b`, so\n `ω'(p)/p = ω_A(p)/p + ω_B(p)/p − ω_A(p)ω_B(p)/p²`, giving\n `Σ_p ω'(p) log p / p = (κ + κ_B) log z + O(1)`. The one-sided `Ω(κ,L)`\n envelope follows from `ω'(p) ≤ ω_A(p) + ω_B(p)`.\n2. **The sifted count multiplies.** `ab` is `z`-rough iff both `a` and `b` are,\n so `S(A',z) = S(A,z)·S(B,z)`, and since `V_{κ+κ_B} = V_κ V_{κ_B}` the\n normalised densities multiply. A factor tending to 0 times a factor bounded\n by 1 tends to 0.\n3. **The level survives.** `|A'_d| = Σ` over the at most `3^{ν(d)}` ways of\n splitting `d` across the two factors, each term `|A_e||B_f|` with `ef | d`.\n So `Σ_{d ≤ y} 4^{ν(d)}|r'_d|` is controlled by the corresponding sums for\n `A` and `B` with a divisor-function weight, which the DHR Theorem 9.1\n remainder condition already carries.\n4. **`s` is unchanged**, because `s = log y / log z` makes no reference to `|A|`.\n\nSo `β(κ + κ_B) ≥ β(κ)` whenever an auxiliary sequence of dimension `κ_B` with\nlevel at least `y` exists, and `B = {m ≤ M}` with `M = y^{1+ε}` supplies\n`κ_B = 1` with a trivial remainder. **Hence `β(κ+1) ≥ β(κ)` and in particular\n`β(2) ≥ β(1) = 2`.** **[DERIVED. Unreviewed. Almost certainly folklore: the\nconstruction is three lines and the padding idea is standard, so this must not\nbe recorded as new, only as located nowhere in the six sources read.]**\n\n**Instantiated, so the hypotheses are visible.** Take `A⁺ = {n ≤ x : λ(n) = +1}`,\nFord §1.7.4 pp. 7–8: `X = x/2`, `g(d) = 1/d`, level `x^{1−o(1)}`, and\n`S(A⁺, √x) = 1`. The level is not a Bombieri–Vinogradov statement: writing\n`n = dm` gives `r_d = (λ(d)/2)L(x/d)` with `L` the Liouville summatory function,\nand the prime number theorem for `λ` alone gives\n`Σ_{d ≤ x^{1−ε}}|r_d| ≪ x/log^A x`. Pad with `B = {m ≤ x}`. Then `A'` has\ndimension 2, level `x^{1−ε}`, `z = √x`, `u = 2 − 2ε`, and normalised sifted\ndensity `≪ log z / x → 0`. So `β(2) ≥ 2`.\n\n**Where it could fail, stated rather than hidden.** (a) The remainder bookkeeping\nin step 3 is sketched, not written out; the `3^{ν(d)}` splitting count and the\ncross terms `|A_e||B_f|` need the two remainder sums to be uniform in the\nsplitting, which is routine but is not checked here. (b) Some conventions tie\nthe level to `|A|` rather than to `z`; under such a convention padding dilutes\n`θ` and the argument gives nothing. Ford's Definition 2 does not, and it is the\ndefinition this note uses. (c) The conclusion is about the class-level `β(κ)`,\nnot about Brady's `inf{s : f_κ(s) > 0}`, which is a property of the DHR\ndelay-differential pair; the two share a symbol and are different objects.\n\n**What it buys, which is little.** `β(2) ≥ 2` is the band's lower endpoint. It\nsays the Zone Postulate's target exponent of 2 sits **exactly at** the axiom\nclass's floor, so no axiom-only argument can go below 2, and an axiom-only\nargument that reached 2 would be attaining the class limit exactly. It says\nnothing about `(2, 4.26645]`.\n\n---\n\n## 3. Sub-question 3: the LP dual, written out, and its finite barrier\n\n### 3.1 The primal, the dual, and what each one is\n\nFix sifting primes `p < z` with `ω(p)` residue classes removed per prime\n(`κ = 2` interval problem: `ω(2) = 1`, `ω(p) = 2` for `2 < p < z`, the classes\nbeing `0` and `−2`). For a sequence `A`, let `y_S` be the density of elements\nwhose prime support below `z` is exactly `S`. Then\n\n- **primal:** `min y_∅` subject to `Σ_{S ⊇ T} y_S = g(T)` for every `T` with\n `∏T ≤ D`, and `y ≥ 0`, where `g(T) = ∏_{p ∈ T} ω(p)/p`.\n\n`y` ranges over non-negative measures on divisor patterns that satisfy every\naxiom the sieve can read at level `D`. **That is precisely the abstract extremal\nproblem the brief describes, and no integer set need realise it.**\n\n- **dual:** `max Σ_{T : ∏T ≤ D} π_T g(T)` subject to `Σ_{T ⊆ S, ∏T ≤ D} π_T ≤ [S = ∅]`\n for every `S`.\n\nSetting `λ_d = π_T` for `d = ∏T`, the dual constraint is exactly the minorant\ncondition `Σ_{d | n, d ≤ D} λ_d ≤ [n = 1]` and the dual objective is exactly the\nmain term `Σ_{d ≤ D} λ_d ω(d)/d`. **So the dual of the abstract extremal problem\nis the general lower-bound sieve, with no restriction to `Λ²` form, to\ncombinatorial weights, or to any construction at all**, and strong duality says\nthe two values coincide: the best main term any level-`D` lower-bound sieve can\ncarry equals the least sifted density consistent with the level-`D` data. A\nvalue of 0 is a barrier for the whole class; a positive value is a sieve.\n\nThis is not new and must not be presented as such: it is the linear-programming\nformulation of the sieve, and Selberg's model problem of §1 is the same object\nwith all primes assumed the same size.\n\n### 3.2 Has anyone computed it at `κ = 2`? Yes, in this corpus, and it was not recorded as a barrier\n\n`research/attack-beta2-04-loss-budget.js` §7 and `research/lp-push-x43.js` solve\nexactly this program at `κ = 1` and `κ = 2` out to `x = 43`, with duality gaps\nto `3.7e-13`, and turn it into the DP1 line of `sift-limit-attack.md` §7d: the\ncalibrated one-point floor **3.3152**, individual readings **2.669 to 3.949**\n(`lp-push-x43.md` §4). In the literature: nothing. `recon-0828-sieve.md` §2\nsearched zbMATH for *sifting limit best possible*, *lower bound for the sifting\nlimit* and *optimality of the combinatorial sieve* and found the only optimality\nstatement in print to be at `κ = 1`. This note adds Halberstam's 2003 review as\nthe authority for that, and Selberg's §17 and Brady's Corollaries as the\nmodel-problem analogue of the LP at large `κ`.\n\n**The point this section exists to make.** `paper/wall-note.md`:429 says \"Inside\n`(2, 4.2665]` there is still no barrier result.\" The corpus's own LP floor at\n`3.3152` **is** a barrier statement inside the band, for exactly the class Face 4\nnames (\"an axiom-only argument\"), and `recon-0828-sieve.md` §5 already uses it as\none: \"Any method reading only `|A_d|` is bounded below by the LP floor 3.3152\".\nFace 4 and that row disagree. The reconciliation is that one is a theorem-level\nclaim and the other is a finite-`x` measurement with a calibration, and Face 4's\nsentence is true of theorems and silent about the measurement it should mention.\n\n### 3.3 Pre-registration, sealed before the companion script was written\n\n*(Written 2026-09-04 before `recon-0904-sifting-limit-floor.js` existed, and\nplaced in this note before the script was written. The script's own output is\nembedded by `research/qc/embed.js`; the scoring sits in section 3.4.)*\n\nThe companion re-implements, in independent code, the level-`D` linear program\nthat `research/lp-push-x43.js` solves, and extends it along one axis that\nprogram never ran: the dimension `kappa` itself.\n\n- **P1, the control.** The independent solver reproduces the twelve published\n raw `s*` cells of `lp-push-x43.js` section 1 (`x = 7, 11, 13, 17, 19, 23` at\n `kappa = 1` and `kappa = 2`) to within `5e-4`, which is the printing\n precision there. A miss on any cell means my implementation, not theirs, and\n nothing below may be quoted.\n- **P2, monotonicity at the LP level.** The raw threshold `s*(kappa)` is\n **strictly increasing** in `kappa` across `kappa in {0.5, 1, 1.5, 2, 3, 4}`\n at every `x in {11, 13, 17, 19, 23}`, with no exception. A single inversion\n is a finding against the LP analogue of \"beta is non-decreasing in kappa\" and\n must be reported as such.\n- **P3, the hard consistency test, and the one that can refute the corpus's\n floor.** DHR's own sieve reads only divisor-class counts, so it lives inside\n the LP's information class, so the calibrated LP reading at any `kappa`\n **must not exceed** the DHR value `beta_kappa` at that `kappa`. Booker and\n Browning's rigorous table gives `beta_3 = 6.64085945080065843931` and\n `beta_4 = 9.07224868172113415960`. Forecast: the `beta_1 = 2` calibrated\n `s*`-route reading lands in `[4.5, 6.0]` at `kappa = 3` and in `[5.5, 8.0]`\n at `kappa = 4`, that is below the DHR value with margin at both. **If any\n calibrated reading exceeds the DHR value at its own `kappa`, the calibration\n that produces the 3.3152 floor is refuted and that floor may not be quoted\n again in this corpus.**\n- **P4, the calibration reproduced.** The calibrated `kappa = 2` reading via\n `s*` at `x = 23` reproduces `lp-push-x43.md` section 3's `3.4889` to within\n `0.01`.\n\n\n### 3.4 Scoring\n\nCompanion: `research/history/staging/recon-0904-sifting-limit-floor.js`, output\nembedded by `research/qc/embed.js`, 458.4 s, 343 LP solves, worst primal\nresidual `1.51e-14`, zero solves failing to certify.\n\n- **P1, the control: PASS.** All twelve published raw `s*` cells reproduced,\n worst disagreement `4.23e-5` against a printing precision of `1e-4`. The two\n solvers share no code.\n- **P2, monotonicity in `κ`: PASS, 25 of 25 strict increases**, zero inversions,\n zero ties, across `κ = 0.5, 1, 1.5, 2, 3, 4` at `x = 11, 13, 17, 19, 23`.\n **[MEASURED]** at five small levels. It is the LP-class analogue of §2's\n derivation and it is not a proof of anything about `β`.\n *A first pass reported one inversion, `s*(1.5) = 6.1308` at `x = 23`. It was a\n solver failure silently read as a zero. The harness now retries under Bland's\n rule and reports failures; the embedded run has none. The first pass's number\n is recorded here because a defect found and fixed is part of the reading.*\n- **P3, the calibration scored against DHR: PASS, and this is the load-bearing\n one.** 0 of 15 calibrated readings exceed the published `β_κ` at their own\n `κ`. Headroom ratio reading/DHR: `0.7442`–`0.8344` at `κ = 2`,\n `0.5813`–`0.7159` at `κ = 3`, `0.4844`–`0.6613` at `κ = 4`. The test could\n have refuted the 3.3152 floor outright and did not.\n- **P4, the published cell: PASS.** `3.4889` at `x = 23` via `s*`, exactly.\n\n**The disconfirming reading inside P3.** The ratio **falls** with `κ`. A floor\nthat tracked the truth would hold its ratio; one that drifts down with `κ` is\na floor that is loosening, which means the instrument is conservative and the\none-point limit could be higher than 3.3152, not lower. That direction is the\none that helps the corpus's inversion claim and hurts any hope of the band.\n\n### 3.5 The barrier made explicit\n\nAt `x = 7`, `κ = 2`, level `D = 21`, `s = 1.5646`, the LP returns a\nnon-negative measure on ten divisor patterns, total mass `1.0000000000`,\nmatching every axiom the sieve can read at that level, with\n`0.000000000000` on the empty pattern. One constraint higher the same program\nreturns `0.004761904762`. So at that level **no argument reading only the\ndivisor-class counts can prove a survivor exists**, and the certificate is a\nten-line table rather than an abstraction. **[VERIFIED]**, and it is a barrier\nfor the method at that finite level, not for the truth: `G₂(7#) = 30` is known\nexactly and the phantom does not contradict it.\n\n---\n\n## 4. Sub-question 4: the best unconditional interval barrier\n\n**Answer: nothing above `β_interval(2) ≥ 3e^{-1/2} = 1.81959`, and that number\nis below 1 only in the sense that it is below 2 and therefore below the band.**\n\n**The route the brief proposes, priced, and it lands at 1.** Take\n`A = {n(n+2) : n ≤ x, λ(n) = +1}`. Two classes per prime, so `ω(p) = 2` and the\ndimension is 2. Level: `#{n ≤ x : λ(n) = +1, n ≡ a mod d}` for `a ≠ 0` is a\nLiouville-in-progressions count, and Bombieri–Vinogradov for `λ` gives level\n`x^{1/2−ε}`, not more. Sifted count: a survivor needs `n` and `n+2` both\n`z`-rough and `λ(n) = +1`, and a `z`-rough integer with `λ = +1` is `1` or\nexceeds `z²`, so the set is empty once `z² ≥ x`, that is `u ≤ 2`. But\n`s = log y / log z = u/2 < 1`. **So the construction gives `β(2) ≥ 1`, which is\ntrivial.** **[DERIVED]** The level is the binding constraint, and it is the same\nconstraint that made the corpus's \"product example, half level\"\n(`sift-limit-attack.md` §2) land outside the operative regime.\n\n**Why the `n ≡ 0` half is not the obstruction.** For `d | n` the count is\n`#{m ≤ x/d : λ(m) = λ(d)}` and the prime number theorem for `λ` alone gives\nlevel `x^{1−o(1)}`, as in §2. It is the second class, `n ≡ −2 mod p`, that needs\n`λ` in a progression to a modulus coprime to `n`, and that is where\nBombieri–Vinogradov caps the level at `1/2`. **The two-class structure is\nexactly what halves the level here**, which is a cleaner statement of the same\nobstruction the corpus records at `sift-limit-attack.md` §2's mixed-sign sketch.\n\n**What would lift it.** An Elliott–Halberstam-strength level for `λ` in\nprogressions, `x^{1−ε}`, would give `s = u < 2` and hence `β(2) ≥ 2`\nunconditionally on intervals. Granville's exceptional character replaces `λ` by\na periodic `χ`, which is why `attack-barrier-kappa2.md` reaches `θ → 1` and\n`β_interval(2) ≥ 2` under a Siegel-zero hypothesis. Neither is unconditional and\nboth stop at 2.\n\n**The only unconditional interval barrier located is method-side, not\nsequence-side.** Brady's model-problem reduction bounds any lower-bound sieve\nregardless of which `A` it is applied to, so `β_interval(2) ≥ 3e^{-1/2}` holds\nunconditionally with no distributional input at all. It is below 2 and therefore\nbelow the band, and it is weaker than §2's padding derivation, which is\nconditional on nothing but is unreviewed. **Neither reaches into `(2, 4.26645]`.**\n\n---\n\n## 5. Sub-question 5: the proposed SEARCH-CONVENTIONS row and the corrections\n\n### 5.1 Proposed §1 row (text only, not applied)\n\n| object | our name | canonical | **OWNING convention, search THIS** | where it lives |\n|---|---|---|---|---|\n| a floor on what any lower-bound sieve of dimension `κ` can do | \"the barrier on Face 4\", \"is 4.2665 a method artefact\" | lower bound on the sifting limit `β_κ` | **Selberg's RECIPROCAL convention `a_k = 1/β_κ`**, in which a lower bound on `β` is titled an **\"upper bound for the sifting limit\"**: search `a_k`, *\"upper bounds for sifting limits\"*, *\"sifting density\"*; and the mechanism convention **\"Selberg's model problem\"**, all sifting primes the same size, parameters `v = Σ_{p∈P} κ/p` and `R`, the quantity to search being **`v_R`**. NOT \"extremal example\", which at `κ > 1` does not exist and is a guaranteed clean negative | Selberg, *Lectures on Sieves*, Collected Papers II (Springer 1991) §§13, 14, 17; Brady, *Sieves and iteration rules*, Stanford PhD 2017, `purl.stanford.edu/gk881hk9239`, Thm 7 and Cor. 1 (p. 11), Thm 22 and Cor. 3 (p. 51), the `v_R` table (p. 46). Ford's 2023 notes Table 1 is captioned \"**Known upper bounds**\" and carries no lower bound. Method-specific limits that must NOT be counted: Grupp–Richert's `ν(κ) > 2κ+1` (Ankeny–Onishi's own), DHR's `α_κ > β_κ > 2` (its own DDE pair) |\n\nA companion line for §3's table: the \"`κ = 2` extremal example\" row's \"None\nfound\" should gain its authority, Halberstam's *Bull. AMS* **40** (2003) p. 116.\n\n### 5.2 Proposed corrections, listed, not applied\n\n1. **`paper/wall-note.md`:422–429, Face 4.** Current: \"no barrier theorem in the\n corpus or in print … **Inside (2, 4.2665] there is still no barrier result.**\"\n Proposed insertion after that sentence: *\"No barrier theorem. The corpus's own\n level-`D` LP does put a measured floor inside the band, at 3.3152 with\n individual readings 2.669 to 3.949 (`lp-push-x43.md` §4), and that floor is a\n statement about exactly this class, so the honest reading of the face is that\n the band is open to a theorem and already contested by a measurement.\"*\n2. **`paper/wall-note.md`:419–421.** Current: \"`β(κ) is unknown in all cases\n κ > 1/2 except κ = 1` … and no such lower bound is known.\" Proposed: *\"…and no\n lower bound on `β(2)` above 2 is known. Lower bounds below 2 are in print:\n Selberg's *Lectures* §17 in his reciprocal convention, and Brady 2017 Cor. 1\n and Cor. 3, giving `β_2 ≥ 1.8394` when his reduction is evaluated at `κ = 2`.\n `β(2) ≥ 2` follows from monotonicity in `κ` by a padding argument\n (`recon-0904-sifting-limit-floor.md` §2), which is unreviewed and probably\n folklore.\"*\n3. **`research/sift-limit-attack.md`:143.** The `[ABSENT — in sieve-theory\n vocabulary]` marker currently covers the whole object. Proposed: keep it for\n the **extremal example**, cite Halberstam 2003 as its authority, and remove it\n for **lower bounds on `β(κ)`**, which are in print under the reciprocal\n convention.\n4. **`research/sift-limit-attack.md`:169.** \"No published example blocks any\n exponent in (2, 4.2665]\" is correct as written; propose adding *\"and no\n published lower bound on `β(2)` reaches 2 either, the best located being\n 1.8394\"*, so the sentence stops being read as \"nothing is known in that\n direction\".\n5. **`research/sift-limit-attack.md` §2, the Franze paragraph.** Add that\n Franze's \"approached rapidly from below\" describes his own Λ²Λ⁻ table, not the\n truth; that his line 272 `\\gtrsim` contradicts his line 73 `\\lesssim` and is an\n apparent typo; that his printed \"`β_κ ≲ 2.44κ` for the DHR sieves\" is the\n growth rate the corpus lacks; and that Ford's notes give the same Selberg\n asymptotic with the constant `0.4454` rather than `19/36 = 0.5278`.\n6. **`research/history/staging/object-g2-read-0829.md`:532–534.** Same edit as\n item 2, since the paragraph reproduces Face 4 verbatim.\n7. **`research/SEARCH-CONVENTIONS.md` §1.** Add the row in 5.1. §1 is parsed at\n run time by the `search-convention` check, so this is a live change to what\n clears the gate and should be reviewed as such.\n\n---\n\n## 6. Provenance\n\n| source | what was read | how | what it carried |\n|---|---|---|---|\n| Booker–Browning ancillary `dhr.html`, arXiv:1511.00601v2/v3 | the `α`, `β` table, full | HTML text layer, `curl` with a browser user agent, and independently by a second fetcher | `β_2 = 4.26645028414864191641`, `β_3 = 6.64085945080065843931`, `β_4 = 9.07224868172113415960` |\n| Franze, arXiv:1012.3809**v1** | introduction, §3, and the LaTeX source `siftinglimitspaper.tex` | PDF text layer via `pdftotext -layout`, plus the e-print source | `β_κ ≲ 2κ + 19/36` for Λ²Λ⁻; \"for the DHR sieves, `β_κ ≲ 2.44κ`\"; the `\\gtrsim`/`\\lesssim` contradiction; no lower bound anywhere |\n| Brady, *Sieves and iteration rules*, Stanford 2017 | pp. 3, 11, 46, 51 and the model-problem chapter | PDF text layer of the Stanford Digital Repository deposit `druid:gk881hk9239`, and a second identical copy at `notzeb.com` | Cor. 1 `β_κ ≥ (1+o(1))2κ/e`; Cor. 3; Thm 22 `v_R ≤ d+2√d+1`; the computed `v_R` table (p. 46) `v_1 = 1`, `v_3 = 2`, `v_5 = 3.11714`; \"It is currently not known whether there is any κ > 1 with β_κ < 2κ\" |\n| Ford, *Sieve Methods Lecture Notes*, `ford126.web.illinois.edu/sieve2023.pdf` | §1.7.4 pp. 7–8, §3.1 p. 37, §4.2 pp. 46–47 | PDF text layer and a rendered page image, agreeing | Definition 2 of `β(κ)`; Table 1 captioned \"Known **upper** bounds\"; `β(κ) ≤ 2κ + 0.4454`; the Liouville example at dimension 1 with level `x^{1−o(1)}` |\n| Blight, Rutgers PhD 2010, RUcore 27420 | printed pp. 6–7, §2.6 pp. 28–29 | PDF text layer, page-isolated | \"a lower bound sieve with a sieving limit of 2κ has not been found for κ > 1\"; `β_κ ∼ cκ`, `c = 3.591…`; upper bounds only |\n| Selberg, *Lectures on Sieves*, Collected Papers II | table of contents p. 65, §14 pp. 165, 179–180, §16 p. 183, §17 p. 201 | **Google Books OCR snippets only**, vol. id `JswcQj2B1HIC`; no page images; fraction bars lost in OCR | §17's title; `a_k = 1/β_κ`; `β_κ = 2κ` as a **conjecture** he himself doubted for large `κ`; §16's examples scoped to `k = 1/2` and `k = 1`; `a_k < e^{1−1/k}/k` |\n| Halberstam, review of Greaves, *Bull. AMS* **40** (2003) | p. 116 | PDF text layer, `curl` with a browser user agent after a 403 to a plain fetch | \"Concrete examples like Selberg's when κ = 1 and Iwaniec's when κ = ½ are not known and greatly to be desired\" |\n| Diamond–Halberstam(–Galway), *A Higher-Dimensional Sieve Method* | pp. 79, 95, 228, index p. 266 | Google Books snippets, vol. id `vnG0S7jMjeYC` | the p. 79 \"trivial lower bound\" sentence verified verbatim; the `Ω(n) mod 2` example stated in the `κ = 1` chapter only |\n| Halberstam–Richert, *Sieve Methods*, Dover reprint | pp. 238–239, 269 | Google Books snippets, vol. id `keKvAAAAQBAJ` | the corpus's p. 239 citation sits in the Notes to Ch. 8, \"The linear sieve\", i.e. `κ = 1`; p. 269 \"When κ > 1 we have at our disposal no result that compares in quality with the main results of Chapter 8\" |\n| Grupp–Richert, *Analysis* **8** (1988) 1–23 | zbMATH review abstract only | zbMATH open API | `ν(κ) > 2κ + 1` for `κ ≥ 7/2`, for the **Ankeny–Onishi** sieve |\n| `research/lp-push-x43.js`, `lp-push-x43.md` | §§0–4 and the embedded output block | in-repo, read | the twelve control cells; the pooled floor 3.3152; readings 2.669–3.949 |\n\n**Channels that failed, exactly.** Selberg's *Collected Papers* Vol. II full\ntext was not reached: `publications.ias.edu/selberg` returns 403 to both a plain\nfetch and `curl` with a desktop Chrome user agent; archive.org's\n`collectedpapers0001selb` is Vol. 1 and lending-restricted; Google Books\n`VgzmnQEACAAJ` (ISBN 9783642410222) has no preview. The working channel was ISBN\n9783540506263, snippet view with search-inside. **Every Selberg quotation in this\nnote is an OCR snippet and none is a page image.** Halberstam–Richert p. 239\nitself was not surfaced: the archive.org item `sievemethods0000halb` is\nlending-restricted and its search-inside endpoint answered 403; about eighteen\ntargeted phrase queries against the Dover volume on Google Books did not reach\nthe page. The Google Books JSON API answers 429 with `quota_limit_value 0`\nwithout a key; the `books.google.dk/books?jscmd=SearchWithinVolume` endpoint\nworks. arXiv's API searches titles and abstracts only, and `all:\"sifting limit\"`\nreturns exactly one paper, Franze.\n\n---\n\n## 7. What this note contradicts in the live layer\n\n1. **`research/sift-limit-attack.md`:143**, the marker `[ABSENT — in sieve-theory\n vocabulary, which is where this object lives]`. **Contradicted for lower\n bounds.** Selberg's *Lectures* §17 and Brady's Corollaries 1 and 3 are lower\n bounds on `β_κ` at `κ > 1`, in print. The absence survives for the extremal\n example only. The search that produced the `[ABSENT]` did read Selberg\n \"through Franze's account\" and did read Brady, so this is not a reach failure\n but a convention failure and a page failure: Brady p. 11 and p. 51 were never\n opened, though pp. 1, 3, 12 and 133 were.\n2. **`research/SEARCH-CONVENTIONS.md`:162**. The row's answer, \"None found. The\n band is a proof gap, not a truth gap\", is right about the example and about\n the band, and its \"what was actually searched\" column lists Brady 2017. It\n should record that Brady contains a lower bound and that the searched\n vocabulary was ours, not Selberg's.\n3. **`paper/wall-note.md`:429**, the sentence \"Inside (2, 4.2665] there is still no barrier\n result.\" **In tension with `recon-0828-sieve.md` §5's last table row**, which\n uses the corpus's own LP floor of 3.3152 as a barrier for exactly this class.\n Both can stand only if \"barrier result\" is read as \"theorem\"; the sentence\n should say so and should name the measurement.\n4. **`research/history/staging/object-g2-read-0829.md`:532–534** reproduces\n Face 4's sentence and inherits both issues.\n5. **`research/sift-limit-attack.md` §2, the Franze paragraph.** \"Two independent\n technologies plateau near 2κ, the community's implicit guess for the\n axiom-class truth at integer κ\" is `[INFERRED]` and is now contradicted by the\n DHR table itself: `β_κ − 2κ` grows without bound over `κ = 2 … 15`, and\n Franze's own printed reading of Chapter 17 is `β_κ ≲ 2.44κ`. The plateau\n holds at small `κ` and the inference does not generalise.\n6. **`research/THE-DIALS.md`:276–279.** The table's `sieve limit β_κ` column\n prints `2` and `4.2665` without saying that the first is a proven two-sided\n value and the second an upper bound only. A footnote is owed.\n\nNothing in this note contradicts `README.md` §Status or `research/G2-STATE.md`\n§0, which were re-read and which do not speak to this face.\n\n---\n\n## 8. What would falsify this note\n\n1. **A published extremal example at `κ = 2`.** It would turn the band from a\n proof gap into a truth gap and would falsify §1's absence. The check that ran\n is Halberstam's 2003 review plus the four theses and note sets of §6; the\n check that has **not** run is the post-2003 literature searched in the\n reciprocal convention, and Selberg's *Lectures* at anything better than OCR\n snippets.\n2. **The Selberg §17 constant being different from `e^{1−1/k}/k`.** Every\n Selberg quotation here is an OCR snippet with the fraction bars lost. The\n direction and scope are corroborated by Brady's sentence; the constant is not.\n A page image of *Lectures* pp. 199–203 settles it and was not obtained.\n3. **An error in the padding argument of §2.** The remainder bookkeeping in step\n 3 is sketched. If the `3^{ν(d)}` splitting cannot be absorbed uniformly, or if\n the intended convention ties the level to `|A|`, `β(2) ≥ 2` falls back to the\n `1.8394` that rests on Brady. Nobody has reviewed step 3.\n4. **The `κ = 2` evaluation of Brady's reduction being outside its hypotheses.**\n His Corollary 3 is stated for `κ ≥ 3`. This note re-runs his own displayed\n chain at `d = 0` and `d = 1`, which requires `v_2 = v_1` and `v_4 = v_3`, an\n identity he asserts inside the proof and this note does not verify.\n5. **The LP calibration failing at a `κ` not tested.** P3 passed at `κ = 3` and\n `κ = 4` at `x ≤ 23`. It has not been run at `κ = 5` or above, where the DHR\n values are larger and the primorial gives the LP less room, nor at `x` beyond\n 23 where the corpus's own solver reaches 43. A calibrated reading above\n `β_κ` at any `(x, κ)` refutes the 3.3152 floor.\n6. **The falling headroom ratio in P3 turning out to be a finite-`x` artefact.**\n If it is, the `κ`-dependence of the calibration is not evidence about the\n floor's tightness in either direction and §3.4's disconfirming reading is\n void.\n", "research/history/staging/rho-exact-z31-01.md": "# Exact Λ and sup|ρ̃| at z = 31 and z = 37: TODO item 0's named cheap data, run\n\n\n\n**Caveats first. Nothing here bounds G₂(z#) and RML(α) is open at every α.**\nWhat ran is two exact points on a measured curve, at a cost of 6928 s, with\nevery control passing. The one verdict-shaped result is a **1.38-standard-error**\nexclusion of β₂ from a seven-point slope fit over less than one octave of z,\nwhich survives neither a two-se bar nor the removal of z = 37, and which cannot\nrefute an asymptotic statement in any case. Four things came out.\n\n**One: the two levels TODO item 0 named, exact (§4.1).** Λ(31) = **476.314**\nand Λ(37) = **1286.338**; sup|ρ̃|(31) = **28.122062** over the complete period\nof 6,469,693,230 positions and sup|ρ̃|(37) = **52.219092** over 200,560,490,130.\nThe first suprema measured past z = 29 in this corpus. ⟨ρ̃²⟩ at both levels\nreproduces the cited column digit for digit by two independent routes, so the\ntwo genuinely new numbers sit on a confirmed lattice.\n\n**Two: MV(α) loses its standing, weakly (§4.2).** The slope moves 4.2144 ±\n0.3039 (5 pts) → **4.5656 ± 0.2173** (7 pts) and β₂ = 4.26645 leaves the 1se bar\nfrom below. Before tonight MV(α) was, in `attack-f4weak-01.md`'s own words, the\nonly form here \"whose truth the measurements do not already decide against\"; it\nnow sits with √N·rms, Λ_V and CH as an instrument measured above β₂ in growth.\nThe exclusion is 1.38 se, inside the 2se bar, and reverses if z = 37 is dropped\n(leave-one-out slope 4.365). It is a MEASURED tension, never a refutation.\n\n**Three, and more useful: the truth has 1.50 exponents of room, and the ℓ¹\naccounting is what spends them (§4.3).** d ln sup|ρ̃| / d ln z = **2.766 ±\n0.212** on seven points against β₂ = 4.26645. The instrument's own loss,\nΛ/sup|ρ̃| running 3.74 → 24.63 across z = 13..37, is an endpoint z^1.802, and\n2.766 + 1.802 = 4.568 recovers the Λ slope to three digits. MV exceeds β₂\nbecause absolute-value accounting over modes bleeds 1.8 exponents, not because\nthe object RML must bound is large.\n\n**Four: the legal law survives at both new levels and the TPC-implying one is\ndestroyed at both (§§4.4–4.5).** The Gaussian ρ law (F4, λ = 0, C = 1) holds\nwith C_true = **0.8412** and **0.7820**, 16% and 22% of margin, extending the\n\"true at every exactly measurable level\" column to seven points. S-sharp fails\non **528 of 640** integer H at z = 31 (worst ratio 7.02) and on **all 26** grid\nvalues to 2048 at z = 37 (worst 15.89). One incidental finding, recorded and not\napplied: the dense scan reproduces `phase1-T4-maximal-law.md` §4's last-failure\nH exactly at z = 13, 17, 19, 23 and finds 201 failures at z = 29 where that note\nrecords **none** — a disagreement with a cited number whose producer was never\ncommitted.\n\n---\n\n## 0. WHAT IS OPEN, FIRST\n\n- **RML(α) is open at every α, and nothing here can close it.** This note buys\n two exact points on a measured curve. A measured slope over seven levels\n spanning less than one octave of z is not a theorem about any exponent, and\n the corpus has already recorded (`rho-maximal-law.md` reading 6,\n `attack-f4weak-01.md` §5) that local slopes in this corpus RISE with the\n window at every instrument looked at. Rung: OPEN, unchanged.\n- **Both new levels are single measurements.** sup|ρ̃| at z = 31 and z = 37 is\n one number each, exact over the complete period to the walker's stated drift,\n with no independent second implementation. The controls below are what stands\n between them and a transcription error, and they are named in advance.\n- **The MV(α) slope is a two-point extension of a five-point fit.** If β₂ leaves\n the bar it is because two points moved a short-window OLS, not because\n anything about the law was proven or refuted. §1 pre-registers what each\n outcome may and may not be said to mean.\n- **z = 41, 43, 47 stay out of reach for the exact sup.** The walk is linear in\n W = P(z), and W = 7.42e12, 3.04e14, 1.31e16 there against 2.01e11 at z = 37:\n 37×, 1517× and 65,232× the level this note reached.\n- **The sharp form's check is dense to z = 31 and a GRID at z = 37.**\n `phase1-T4-maximal-law.md` §4 scanned integer H densely and reports \"last H\n at which S-sharp fails\". This note scans EVERY integer H in 1..640 wherever that is affordable\n (z ≤ 31, which contains T4's 451-value scan and the windows nP = 60..588), and\n falls back to a 26-value grid to 2048 at z = 37, where the dense scan prices\n at ~12 h. A grid that finds no failure has not proven S-sharp holds at every\n H; the z = 37 row is therefore weaker evidence than the others and says so.\n The z = 13..29 rows are carried precisely so the scan can be judged against\n T4's recorded failures at 1, 35, 227, 197.\n\n## 1. PRE-REGISTRATION (written before Λ(31) and Λ(37) existed)\n\n**State of knowledge at the moment this section was written.** Λ(z) =\nΣ_{e,a}|Θ_e(a)|/(2 sin(πa/e)) is exact in the corpus at five levels only:\n9.801, 25.157, 60.031, 121.250, 261.555 at z = 13, 17, 19, 23, 29\n(`attack-f4weak-01.md` §2/S3), OLS slope d lnΛ/d lnz = 4.214 ± 0.304 against\nβ₂ = 4.26645, β₂ inside the bar. Λ(31) and Λ(37) have never been computed; the\nonly figures for them in the corpus are the FLAGGED constant-share model\nΛ-hat = 4.369e+2 and 1.101e+3, which `attack-f4weak-01.md` states leans LOW.\nThis session has NOT computed Λ at 31 or 37 at the time of writing.\n\n**Disclosed, so the blindness claim is exact:** scratchpad pricing probes for\nthis note walked PARTIAL periods and so already produced lower bounds on\nsup|ρ̃| — ≥ 24.7168 at z = 31 (5% of the period) and ≥ 43.4377 at z = 37 (1%).\nThe sup pre-registration below is therefore only partly blind and is written\nknowing those two floors. The Λ pre-registration is fully blind.\n\n**Outcome A — β₂ stays inside the seven-point bar.** Reading: the measured\ngrowth of the ℓ¹ mean value over modes remains statistically indistinguishable\nfrom the sifting-limit exponent on the available window. What follows: nothing\nabout MV(α)'s truth. What does NOT follow: any claim that MV(α < β₂) is\nplausible, likely, or supported. Rung: MEASURED, seven points, one octave.\n\n**Outcome B — the bar excludes β₂ from ABOVE (slope + 1se < β₂).** Reading: a\nmeasured tension in MV(α)'s FAVOUR — the weakest sufficient form's own growth\nsits below the proven sieve limit on this window. What follows: the object\nworth an analytic attempt is named and its target exponent is quantified. What\ndoes NOT follow: any exponent for G₂(z#). A slope is not a bound; MV(α)\nrequires a uniform constant C and a proof for all z, and a fitted slope over\nz ≤ 37 supplies neither. Rung: MEASURED tension, never evidence for a theorem.\n\n**Outcome C — the bar excludes β₂ from BELOW (slope − 1se > β₂).** Reading: a\nmeasured tension AGAINST the weakest sufficient form. What follows: MV(α) joins\nthe instruments this corpus has already watched die by exceeding β₂ in growth\n(√N·rms at 8.174 ± 0.249, Λ_V at 5.230 ± 0.186, CH at 5.214 ± 0.151). What does\nNOT follow: refutation. MV(α) is an asymptotic statement; a finite-z slope that\nexceeds β₂ over one octave does not contradict it, exactly as the Bonferroni\nx ≤ 227 and the trivial-ℓ¹-suffices-to-z~64 phenomena in this corpus do not.\nRung: MEASURED tension, never a refutation.\n\n**In every outcome the standing rider holds:** local slopes in this corpus rise\nwith the window at every instrument measured so far, so a seven-point slope is\nan upper-biased estimate of nothing in particular and quoting a single power is\nrefused (`attack-f4weak-01.md` §5's own instruction).\n\n**Pre-registered on the Gaussian ρ law** (F4 at λ = 0, C = 1: sup|ρ̃| ≤\n√⟨ρ̃²⟩·√(2 lnW)). Measured TRUE at all five exact levels with C_true =\n0.6362, 0.6044, 0.8022, 0.7554, 0.7830. If C_true < 1 at 31 and 37, the law\nsurvives two more levels and the C_true column becomes seven points; that is\nMEASURED and is not evidence the law holds for all z. If C_true > 1 at either,\nthe legal law is FALSE as literally stated at a computable level, which would\nbe the first such reading and would demote F4 from \"measured true everywhere\ncheckable\" to \"false at z = 31 (or 37)\" — a real correction to\n`rho-maximal-law.md` reading 3 and to `rho2-analytic-bound.md` §4's F4 row.\n\n**Pre-registered on the sharp form** (S-sharp: sup_x|R_H(x)| ≤\n√(2 lnW)·√(⟨R²⟩_H), TPC-implying, violated by 0.6% at z = 19, H = 191). If\nS-sharp fails somewhere on the H grid at z = 31 or 37, the \"already false as\nliterally stated\" reading extends past z = 23. If it holds on the whole grid at\nboth, that is a NULL on a coarse grid and says nothing beyond the grid, since\nT4 already records that z = 29 has no failure at any of 451 scanned H.\n\n## 2. THE PRICE, WRITTEN BEFORE THE RUN\n\nPriced from measured neighbours, not from guesses, per the standing compute\nrule. Every figure below is a scratchpad measurement on this machine\n(10 cores, 8 performance) taken before the producer was written:\n\n| piece | measured basis | price |\n|---|---|---|\n| exact Λ, z = 29 | re-ran `attack-f4weak-01.js`'s `l1exact` verbatim: 261.555 in **4.9 s** | — |\n| exact Λ, z = 31 | inner work Σ_e φ(e)·\\|m1\\|/2 = 4.31e8 vs 1.15e8 at z = 29, ×3.75 | **~18 s** |\n| exact Λ, z = 37 | inner work 3.55e9, ×31 over z = 29 | **~155 s** |\n| full-period walk, z = 29, 26-value H grid | measured, 8 workers | **6 s** |\n| full-period walk, z = 29, dense H = 1..640 | measured | **57 s** |\n| full-period walk, z = 31, dense H = 1..640 | measured on 2% of the period, ×50 | **~1750 s (29 min)** |\n| full-period walk, z = 37, 26-value H grid | measured on 1% of the period, ×100 | **~4600 s (77 min)** |\n| closed-form ⟨ρ̃²⟩ and ⟨R²⟩_H | measured: 9.0 s at 29, 28.4 s at 31, O(n²) → ~130 s at 37 | **~180 s** |\n| **total** | | **~113 min, under the 4 h ceiling** |\n\n**What the engine could NOT reach, and what would be needed.** The single-thread\nwalker of `attack-f4weak-01.js` prices z = 37 at 899 × 17.2 s ≈ 4.3 h before any\nH tracking, i.e. OVER the ceiling; that is why `rho-maximal-law.md` §7 lists\nonly z = 31 as the next exact point. What buys z = 37 is not a new\nnormalisation but the same walk cut into re-seeded chunks: ρ̃(a) is computable\nexactly at any a in O(n) from the terms, so the period splits into independent\nblocks with no accumulated drift across them, and the block width was dropped\nfrom 2²² to 2²⁰ so the stamping array is L2-resident (measured 2.2× on its own,\nbefore any parallelism). Together: 4.3 h → 75 min. **z = 47 remains out of\nreach by the same accounting: W = 1.31e16 is 65,232 times z = 37's period, so\nthe exact sup there is ~10⁵ core-hours on this engine and no reorganisation of\nit changes that; it needs a different object, not a faster walk.**\n\n## 3. DEFINITIONS (unchanged from the existing exact levels)\n\ns = 3.0, D = z^s, terms from `sift-limit-lemmaV.js`'s `buildTerms(z, D)` —\nidentical call, identical lattice, identical weights. ρ(y) = Σ_j w_j ψ((y−c_j)/q_j),\nψ(t) = t − ⌊t⌋ − 1/2, period W = P(z) = ∏_{p exact max drift\n 31 6,469,693,230 28.122062 476.314 24.738489 9.9e-8\n 37 200,560,490,130 52.219092 1286.338 85.677253 2.5e-7\n```\n\n⟨ρ̃²⟩ at both levels reproduces the cited column digit for digit (24.738487 in\n`attack-rhoms-01.md` S1; 9.25620² = 85.6772 from `theta-ladder.md` §2), by two\nindependent routes here — Parseval over the modes and the O(n²) closed form —\nso the two genuinely new numbers, sup|ρ̃| and Λ, sit on a lattice whose second\nmoment is confirmed three ways.\n\n### 4.2 MV(α): the slope, and what it does to β₂\n\n```\n window slope 1se 1se bar beta_2 2se bar\n 5 pts, 13..29 (standing) 4.2144 0.3039 [3.9105, 4.5183] INSIDE [3.6066, 4.8223] INSIDE\n 6 pts, 13..31 4.3651 0.2411 [4.1239, 4.6062] INSIDE [3.8828, 4.8473] INSIDE\n 7 pts, 13..37 (this file) 4.5656 0.2173 [4.3482, 4.7829] OUTSIDE [4.1309, 5.0002] INSIDE\n```\n\n**This is pre-registered Outcome C, and it is weak.** β₂ = 4.26645 leaves the\none-standard-error bar from BELOW, by **1.38 standard errors** — it stays\ncomfortably inside the two-se bar. Two further reasons to hold it lightly, both\nin the producer's own output:\n\n- **The exclusion is not robust to dropping one level.** Leave-one-out slopes on\n the seven points read 4.701, 4.476, 4.613, 4.566, 4.672, 4.561, **4.365**;\n the last is the fit with z = 37 removed, and 4.365 puts β₂ back inside.\n One level carries the sign of the verdict.\n- **The local slopes are not a power law and never have been.** 3.51, 7.82,\n 3.68, 3.32, **8.99**, **5.62** across the six consecutive gaps. The 29 → 31\n local slope of 8.99 is the largest in the corpus for this object; ln(31/29) is\n a 0.067 lever, so a single level's jump reads as an enormous local exponent.\n `attack-f4weak-01.md` §5's instruction to quote no single power stands.\n\n**One prior prediction scored.** `attack-f4weak-01.md` S3's flagged\nconstant-share model Λ-hat put z = 31 at 4.369e+2 and z = 37 at 1.101e+3, with\nits own rider that the falling a ≤ 3 share makes it lean LOW. Measured: 476.314\nand 1286.338, so the truth came in **9.0% and 16.8% above the model** and its\nrider was right in direction and roughly right in size. That is a scored forecast, not a\ncontrol, and it is recorded because the model is still cited at z = 41..47 where\nnothing has been measured: it should be read as a lower bound drifting further\nlow with z, and the 5.649 ± 0.227 slope that model carries at 31..47 is\ncorrespondingly biased.\n\nPer §1's Outcome C: this is a MEASURED tension against the weakest sufficient\nform, on a one-octave window, and it is **not a refutation**. MV(α) is an\nasymptotic statement; a finite-z slope above β₂ over z ≤ 37 contradicts nothing,\nand this corpus has watched the trivial ℓ¹ law \"suffice\" to z ~ 64 and the\nBonferroni bound \"hold\" to x ≤ 227 for exactly this reason.\n\n**What DID change, and it is the substantive reading.** Before tonight MV(α) was\n\"the only form here whose truth the measurements do not already decide against\"\n(`attack-f4weak-01.md` §2). It no longer holds that position: it now sits with\n√N·rms (8.174 ± 0.249), Λ_V (5.230 ± 0.186) and CH (5.214 ± 0.151) as an\ninstrument measured ABOVE β₂ in growth, differing from them only in how far\nabove and in how weak the exclusion is.\n\n### 4.3 Why MV moved: the ℓ¹ instrument's loss, not the truth\n\nThe truth and the instrument are now both measured at seven levels, and they\nseparate:\n\n```\n d ln sup|rho~| / d ln z = 2.497 +/- 0.346 (5 pts) -> 2.766 +/- 0.212 (7 pts)\n d ln Lambda / d ln z = 4.214 +/- 0.304 (5 pts) -> 4.566 +/- 0.217 (7 pts)\n Lambda / sup|rho~| = 3.74, 5.80, 6.56, 10.02, 14.61, 16.94, 24.63 at z = 13..37\n```\n\nThe last row is arithmetic on the two measured columns. Its endpoint ratio is\nz^1.802 across z = 13 → 37 (two-endpoint, NOT a fit), and 2.766 + 1.802 = 4.568\nrecovers the Λ slope to three digits. **So the ℓ¹ mean value exceeds β₂ because\nof its own accumulating loss against the truth, not because the object it bounds\nis growing fast.** The object RML(α) actually has to bound grows at **2.766 ±\n0.212**, which is **1.50 exponents below β₂** and leaves the pricing lemma\nenormous room. Rung: MEASURED, seven levels, one octave; the same short-window\nrider applies to the 2.766 as to the 4.566, and the two riders do not cancel.\n\n### 4.4 The Gaussian ρ law at the new levels: still true\n\n```\n z sup|rho~| sqrt sqrt(2 lnW) C_true alpha_true theta_true theta_G\n 31 28.122062 4.97378 6.7217 0.8412 0.9716 2.2071 2.25232\n 37 52.219092 9.25620 7.2144 0.7820 1.0954 2.2797 2.34513\n```\n\n**F4 at λ = 0, C = 1 holds at both new levels**, with 16% and 22% of margin.\nThe C_true column is now seven points: 0.6362, 0.6044, 0.8022, 0.7554, 0.7830,\n**0.8412, 0.7820**. z = 31 is the highest value the column has ever taken and\nz = 37 falls back below z = 29, so the column is NOT monotone and the fitted\ndrift (0.255 ± 0.099 per ln z, crossing C_true = 1 at z ~ 75) is a straight line\nthrough a non-monotone seven-point column. **That crossing is a fit artefact\nquoted for scale, not a prediction that the law breaks near z = 75**, and no\nweight should be put on it; the corpus has no mechanism that says C_true must\ndrift at all.\n\nθ_true, the exponent the truth actually delivers under the strict T ≥ 1\naccounting, is 2.2071 and 2.2797 — BELOW the Gaussian law's θ_G at the same\nlevels (2.25232, 2.34513), by 0.045 and 0.065 of exponent, which is exactly\nC_true < 1 restated. And sup|ρ̃| ≤ smax holds at both levels with room:\n28.12 ≤ 3.372e+4 and 52.22 ≤ 6.881e+4.\n\n### 4.5 S-sharp: dead at both new levels, and a candidate correction at z = 29\n\n```\n z scan #H failing worst ratio (at H) LAST H failing T4 sec.4 records\n 13 dense 1..640 1/640 1.1147 (H=1) 1 1\n 17 dense 1..640 35/640 2.4812 (H=3) 35 35\n 19 dense 1..640 153/640 3.3245 (H=3) 227 227\n 23 dense 1..640 90/640 3.0812 (H=3) 197 197\n 29 dense 1..640 201/640 2.9648 (H=3) 221 none\n 31 dense 1..640 528/640 7.0169 (H=1) 629 —\n 37 grid to 2048 26/26 15.8889 (H=1) 2048 —\n```\n\n**At z = 31 and z = 37 the TPC-implying sharp form is not \"transiently violated\"\n— it is broken across the whole scan.** At z = 31, 528 of 640 integer H fail and\nthe worst ratio is 7.02; at z = 37 every one of the 26 grid values fails,\nincluding H = 2048, with a worst ratio of 15.89. Reading: S-sharp as literally\nstated is FALSE at these levels, which is what `phase1-T4-maximal-law.md` §5\nalready concluded structurally (a form that would imply TPC cannot be a\ngenerous hypothesis) and what its z = 19 row showed at 0.6%. Rung: MEASURED,\nexact sup over complete periods. What does NOT follow: nothing about the LEGAL\nρ law, which forfeits exactly the lag cancellation S-sharp keeps; §4.4 above is\nthe relevant column and it holds.\n\n**Two caveats on the two new rows, both structural.** At z = 31 the last failing\nH, 629, sits 11 short of the top of the dense range, so \"last H failing\" there\nis a lower bound on the truth, not the truth. At z = 37 the scan is a 26-value\ngrid and finding all 26 failing is not the same statement as a dense scan; the\nnumber to quote is \"every grid value to 2048 fails\", not \"every H fails\".\n\n**The candidate correction.** The dense scan reproduces T4 §4's last-failure H\nEXACTLY at z = 13, 17, 19 and 23 (1, 35, 227, 197 — four for four, an engine\ncalibration against a known failure, not a null). At z = 29 it finds 201 failing\nH with the largest at 221, where T4 records **none**. Given four-for-four\nagreement below it, the most likely reading is that T4's z = 29 row is wrong or\nthat its scan range at z = 29 did not include H ≈ 221. It cannot be diagnosed\nfrom here: `phase1-T4-maximal-law.md` §0 lists its producers (`t4-sup.js` et al.)\nas **uncommitted scratch scripts**, so there is nothing to re-run. Rung: a\nMEASURED disagreement with a cited number that has no reproducible producer.\nThis note does not edit that file; the discrepancy is recorded here for\nadjudication.\n\n### 4.6 The second moment against the proven bound: nothing moves\n\nThe bound/exact ratio table now carries own exact values at 31 and 37 in place\nof cited ones:\n\n```\n z FORM I* I*/exact R1f/smax (cited) (3W)^{1/3}/smax (cited)\n 29 13.595974 2.3728e+15 1.745e+14 0.076 0.076 0.032 0.032\n 31 24.738489 4.0882e+15 1.653e+14 0.232 0.232 0.080 0.080\n 37 85.677253 1.7123e+16 1.999e+14 0.541 0.541 0.123 0.123\n 47 360.901967 1.1597e+17 3.213e+14 15.343 15.346 2.155 2.155\n```\n\nEvery cited figure reproduces, including the corrected absolute-floor row from\n`redteam-0828-closures.js` R5 (clears at 41 and 43, misses at 47 by 2.155×) and\nthe I*/exact endpoint control at z = 13 (2.535e+12, MATCH). The non-monotonicity\n`rho2-analytic-bound.md` §0 flags is confirmed on own numbers: 1.745e+14 at\nz = 29 EXCEEDS 1.653e+14 at z = 31. **Nothing here moves any exponent**, and the\nstanding verdict is unchanged: the mean-square lemma is an ingredient of RML,\nnever a route, because Chebyshev off any second moment cannot reach a sup law.\n\n## 5. READINGS\n\n1. **MEASURED, new.** Exact Λ at the two levels TODO 0 named:\n **Λ(31) = 476.314, Λ(37) = 1286.338**, each Parseval-controlled against the\n PROVEN closed-form ⟨ρ̃²⟩ (rel 1.8e-11 and 6.0e-10) and produced by\n `attack-f4weak-01.js`'s recursion unaltered, which reproduces all five cited\n values at the levels where they exist. Producer S1.\n\n2. **MEASURED, new.** Exact sup|ρ̃| over complete periods:\n **28.122062 at z = 31** (W = 6.47e9) and **52.219092 at z = 37**\n (W = 2.01e11), the first supremum measured past z = 29 in this corpus.\n Period mean 0 to 1.2e-7, walked m2 against the closed form to 3.8e-12, worst\n chunk drift 2.5e-7. Producer S2.\n\n3. **MEASURED tension, pre-registered Outcome C, WEAK.** MV(α)'s slope moves\n 4.2144 ± 0.3039 → **4.5656 ± 0.2173**, and β₂ = 4.26645 leaves the 1se bar\n from below by **1.38 se**. It stays inside the 2se bar, and dropping z = 37\n alone returns 4.365, which puts it back inside at 1se. What this is: the\n weakest sufficient form is no longer the only instrument the measurements do\n not decide against. What this is NOT: a refutation of MV(α), which is\n asymptotic and untouched by any finite-z slope. §4.2.\n\n4. **MEASURED, and the more useful half of reading 3.** The truth's own growth\n is **d ln sup|ρ̃| / d ln z = 2.766 ± 0.212** on seven points, **1.50\n exponents below β₂**. The ℓ¹ instrument's slope is the truth's plus its own\n measured loss: Λ/sup runs 3.74 → 24.63 across z = 13..37, an endpoint z^1.802,\n and 2.766 + 1.802 = 4.568 recovers the Λ slope. MV exceeds β₂ because the ℓ¹\n accounting bleeds 1.8 exponents, not because the object is large. §4.3.\n\n5. **MEASURED.** The Gaussian ρ law (F4, λ = 0, C = 1) **holds at both new\n levels**: C_true = 0.8412 (z = 31) and 0.7820 (z = 37), margins 16% and 22%.\n Seven-point column, non-monotone; the fitted crossing at z ~ 75 is an\n artefact of fitting a line to a non-monotone column and is quoted for scale\n only. §4.4.\n\n6. **MEASURED.** S-sharp, the TPC-implying form, is **broken across the whole\n scan at both new levels**: 528 of 640 integer H fail at z = 31 (worst ratio\n 7.02) and all 26 grid values to 2048 fail at z = 37 (worst 15.89). The\n asymmetry `rho-maximal-law.md` reading 3 records — the legal law true, the\n TPC-implying law false — widens rather than closes at the new levels. §4.5.\n\n7. **MEASURED disagreement with an uncustodied cited number.** The dense scan\n reproduces T4 §4's last-failure H exactly at z = 13, 17, 19, 23 and finds\n 201 failing H at z = 29 with the largest at 221, where T4 records \"none\".\n T4's producers are uncommitted scratch scripts, so this cannot be diagnosed\n from the repository. Recorded, not applied. §4.5.\n\n8. **MEASURED, confirmatory only.** ⟨ρ̃²⟩ at 31 and 37 reproduces the cited\n column digit for digit by two independent routes; the FORM I*/exact ratio\n table, the R1f floor and the corrected absolute floor all reproduce, and the\n ratio's recorded non-monotonicity (1.745e+14 at 29 > 1.653e+14 at 31) is\n confirmed on own numbers. Nothing about any exponent moved. §4.6.\n\n9. **ENGINE, auditable.** z = 37's exact supremum was reached inside the compute\n ceiling by chunking the same walk with exact O(n) re-seeding and an\n L2-resident Int32 stamping block, not by changing any definition; the old\n single-thread walker and the new one agree at z = 19 and z = 29 to inside the\n old one's own drift. z = 47 stays out of reach by a factor 3.1e6 in W and no\n reorganisation of this engine reaches it. §2.\n\n## 6. NOT REACHED\n\n- **sup|ρ̃| at z = 41, 43, 47.** W = 7.42e12, 3.04e14, 1.31e16 against 2.01e11\n at z = 37, and the walk is linear in W: **z = 41 prices at ~48 h** on this box\n (37 × the 77-minute z = 37 walk, over the 4 h ceiling but not absurd),\n z = 43 at ~81 days, z = 47 at ~9 years. These need a different object, not a\n faster walk.\n- **A dense H scan at z = 37**, priced at ~12 h; the 26-value grid stands in for\n it and the row is labelled accordingly.\n- **The last failing H at z = 31 beyond 640.** 528 of 640 fail and the largest\n failure is at 629, so the recorded value is a lower bound.\n- **Any adjudication of the z = 29 S-sharp disagreement**, which needs a decision\n about a cited number whose producer was never committed.\n- **The s-freedom.** Everything here is s = 3.0, as at every existing exact level.\n\n## 7. CUSTODY AND GATE\n\nEvery measured figure in this note is in `research/rho-exact-z31-01.js`'s OUTPUT\nblock, written by `node research/qc/embed.js --timeout 14000` — the tool ran the\nscript and wrote the block, so nothing was transcribed by hand\n(code-sha256 `433689f5…`, out-sha256 `b0519fcd…`, 108 body lines, elapsed\n6928.4 s, embedded 2026-08-28). A `qc/embed.js --check` re-run costs another\n115 minutes and was not bought; the block is bound by construction because the\ntool produced it in the same invocation that ran the code.\n\n**Derived arithmetic in this note, done on the embedded columns and not by the\nscript:** the Λ/sup|ρ̃| row (3.74, 5.80, 6.56, 10.02, 14.61, 16.94, 24.63), its\ntwo-endpoint exponent 1.8020, and the 1.377 standard errors by which β₂ sits\nbelow the seven-point slope. Nothing else in §§4–5 is computed outside the\nproducer.\n\n**Cited, custody-bound, reproduced in-pass rather than trusted:** Λ at\nz = 13..29 (`attack-f4weak-01.md` S3), sup|ρ̃| at z = 13..29 (`rho-maxlaw-01`\nS1), θ_G at all ten levels (`theta-ladder.md` §2 via `rho-maxlaw-01` S3, worst\n|recomputed − cited| = 4.2e-5), C_true at z = 13..29, rmsr at 41..47, the\nFORM I*/exact endpoint (`rho2-analytic-bound.md` S1), both Chebyshev floor rows\nincluding `redteam-0828-closures.js` R5's correction, and T4 §4's last-failure H\ncolumn. β₂ = 4.26645 per `paper/beta2-note.md`.\n\n**Gate: `node research/qc.js` was NOT run** (out of scope for this pass by\ninstruction). No existing file was edited by this session: the producer and this\nnote are both new, and the two things this note argues bear on other files —\nT4 §4's z = 29 row and MV(α)'s standing in `attack-f4weak-01.md` and TODO 0 —\nare recorded here for adjudication and applied nowhere.\n", "research/history/staging/rho-maximal-law.md": "# The rho maximal law: the statement, the sufficiency curve, the literature position\n\n\n\n**TODO item 0's named first move, executed: the law is stated, its weakest\nsufficient form is priced against the (31, 47] crossing, and the position\nagainst print is settled. Producer: `research/rho-maxlaw-01-sufficiency.js`\n(all controls pass; output embedded). Four things came out.**\n\n**One: the statement, and a ladder of five forms from tautology to Gaussian\n(§1).** The clean absolute form is RML(α): `sup_y |ρ̃_z(y)| ≤ C·z^α` uniformly\nin z, for the H-free sawtooth potential of the Brüdern–Fouvry certificate at\ns = 3.0. The pricing lemma (§2) turns any RML(α) with α < β₂ = 4.26645 into\n`G₂(z#) ≪ z^α ln²z` — an unconditional two-class exponent below 4.2665, the\nfirst movement of β₂ since 2008. Every step of the lemma is a proven identity\nor standard linear-sieve asymptotics except the law itself; the T ≥ 1\naccounting demanded by the fixed-point rider is built in and costs 1/M =\nO(ln²z), i.e. at most 0.045 of exponent at z = 13 falling to 0.0008 at z = 47.\n\n**Two: the sufficiency curve (§3).** In Gaussian units the allowed looseness is\n`λ_max(z) = β₂ − θ_G(z)`, measured 2.31949 at z = 13 falling to **1.78446 at\nz = 47, the weakest measured point**: a maximal law weaker than the Gaussian\nform by a factor z^1.78 — about 963 at z = 47 — still beats β₂. The curve\nfalls at the crude column's drift rate (0.3909 ± 0.0363 per ln z over ten\npoints); whether it ever reaches zero is model-dependent and undecided — the\nlinear-drift model closes the room at z ~ 5·10³, the power-law model leaves\n1.53 ∓ 0.14 of exponent forever, and ten points over one octave cannot\nseparate them (§6, neither called). The (31, 47] crossing raises the FLOOR\npast the zone budget (th(nP) ≥ 2.0106 at z = 47) and kills only the\nTPC-grade deliverable: against β₂ it consumes 0.5% of the room, leaving a\nband 2.256 exponents wide for a provable law to land in.\n\n**Three, a new measurement: the Gaussian ρ law is TRUE at every level it can\nbe checked, with 20–40% of margin.** `C_true = sup|ρ̃|/(√⟨ρ²⟩·√(2 lnW))` =\n0.6362, 0.6044, 0.8022, 0.7554, 0.7830 at z = 13..29, exact over complete\nperiods — the first direct measurement of sup|ρ| in the corpus (everything\nprior measured sup|R_H|). Unlike the sharp R_H law, which is TPC-implying and\nalready violated by 0.6% at z = 19, the ρ law is legal, looser, and so far\ntrue.\n\n**Four, the literature verdict: the sufficiency curve's weakest point is ABOVE\npublished strength — no headline, the theorem is genuinely open.** The\nstrongest published instrument for this remainder (Bettin–Chandee 2015,\ntrilinear Kloosterman fractions, priced 2026-08-19 at our configuration)\ndelivers exponent 5.0907, i.e. 0.824 of exponent ABOVE the 4.26645 ceiling\n(θ_total 1.0303 in print against 1.2090 needed). Nothing in print bounds the\ntwo-class object at any exponent (ABSENT, owning convention, re-verified\n2026-08-18). One held reframing finding: at the sufficiency ceiling the\nworking point is s/u = 0.703, INSIDE Lemma V's stated range — the β₂ target,\nunlike the TPC target, is Lemma-V territory (§5, HELD).\n\n---\n\n## 0. CUSTODY\n\nEverything measured tonight is in `research/rho-maxlaw-01-sufficiency.js`\n(embedded tail, `qc/embed.js --check`: code-sha256 matches, out-sha256\nmatches). Division of labour, per the standing compute rule:\n\n- **Recomputed** (cheap): the conditional Gaussian-ρ column at z = 13..31 from\n the repo's own `row(z, ·, 3.0)` — all six rows MATCH the embedded tables\n digit for digit at printed precision (`theta-ladder.md` §2 at z = 19..31;\n `phase1-T4-maximal-law.md` §2 at z = 13, 17). The column is H-free and is\n NOT touched by the u = 3.2 defect, which lives only in §5b's old\n `need_true`; nothing from the refuted §5b column is used anywhere here.\n- **Cited** (expensive, custody-bound): the z = 37..47 conditional rows from\n `theta-ladder.md` §2 (independently reproduced by T4's custody block;\n re-running them is ~2800 s at z = 47). Their cheap fields (n, M, lnW) were\n recomputed from `buildTerms` and MATCH exactly.\n- **Cited**: the corrected self-consistent windows `nP(z)` from\n `history/staging/theta-selfconsistent.md` (exact 60, 126, 198, 258, 390,\n 588 at z = 13..31; prefix lower bound 2301 at z = 47), post-rewrite\n instrument only.\n- **New measurement**: sup|ρ̃| exact over complete periods at z = 13..29.\n Walker controls in-pass: brute force at z = 13 (identical to 1e-8), period\n mean = −M/2 at every z, walked variance against the PROVEN closed-form\n ⟨ρ²⟩ to 1.2e-8 %, period-closure drift ≤ 4.3e-7.\n- **Estimator control**: the OLS fitter returns slope 2.500000 on a known\n z^2.5 truth at the same ten z before any fit is quoted, and the nine-point\n θ_G drift refit reproduces T4 §2b's 0.3746 ± 0.0404 as 0.3747 ± 0.0405.\n\nβ₂ = 4.26645 per `paper/beta2-note.md` / `SEARCH-CONVENTIONS.md` §4 (four-\ndecimal rounding of Diamond–Halberstam's twenty-decimal value; unattributed\nper the citation discipline there). Units: windows in positions (integers);\nevery comparison of strengths is stated in exponents base z, dimensionless.\n\n---\n\n## 1. THE STATEMENT\n\n**The object.** Fix the family parameter s = 3.0 (any s > 1+√e = 2.6487 gives\na positive main term; 3.0 is the measured family). For a prime z let\nW = P(z) = ∏_{p 0 the position-free main density. The\n**potential** is\n\n ρ_z(y) = Σ_j w_j · ψ((y − c_j)/q_j), ψ(t) = t − ⌊t⌋ − 1/2,\n\nperiodic mod W with exact period mean −M/2 (each ψ term averages −1/(2q));\nwrite ρ̃ = ρ + M/2 for the centered potential, and ⟨ρ²⟩ for its period\nvariance, which has a PROVEN closed form (`sift-limit-lemmaV.js` S2). The\ncertificate identity, also proven, is T(x) = H·M + R_H(x) with\nR_H(x) = ρ(x) − ρ(x+H).\n\n**The law — five forms, weakest sufficient upward** (each implies the win of\n§2 when its parameter clears the stated threshold):\n\n- **F0 (tautological floor, not a law):** nP(z) ≤ z^{β₂−ε}. This is the win\n restated; pricing stops being informative here. Listed so the ladder has\n its bottom.\n- **F1 (one-sided, single-window):** ∃ε > 0, z₀: for every z ≥ z₀ there is an\n H ≤ z^{β₂−ε} with min_x R_H(x) ≥ 1 − H·M. Only the lower deviation of R_H\n matters; upper deviations are harmless. This is the literal weakest\n law-shaped form.\n- **F2 (sup-norm Lemma V at the ceiling):** at s = 3.0, for H ≍ z^{β₂−ε}:\n sup_x |R_H(x)| ≤ H·M − 1. Note s/u = 3/4.26645 = 0.703, INSIDE Lemma V's\n stated range s ≥ (0.63+δ)u (see §5; every delivered exponent in (3, β₂)\n keeps the working point inside that range, and only exponents ≤ 3 leave\n it). Lemma V's own conclusion ≪ H/log³H is strictly stronger than needed.\n- **F3 (the ρ maximal law, absolute form — THE NAMED TARGET):** **RML(α):\n ∃C, z₀ such that for every prime z ≥ z₀, sup_{y∈ℤ/W} |ρ̃_z(y)| ≤ C·z^α.**\n Sufficient whenever α < β₂ = 4.26645. Via |R_H| ≤ 2 sup|ρ̃| this implies F2\n and F1 at every window and needs no per-H information.\n- **F4 (Gaussian-units form):** RML_G(λ, C): sup|ρ̃_z| ≤ C·z^λ·√⟨ρ²⟩·√(2 lnW).\n λ = 0, C = 1 is the corpus's Gaussian maximal law\n (`sift-limit-lemmaV.js`:466), the form measured TRUE at z = 13..29 (§4).\n Sufficient whenever λ < λ_max(z) uniformly (the §3 curve).\n\nQuantifier discipline: the sup is over ALL W positions of the complete period\n(equivalently all y ∈ ℤ); one constant uniform in z; s fixed. **What this law\nis NOT:** the sharp maximal law for R_H at the operative (~0.5 z²) lag, which\nis TPC-implying and therefore not a legal target\n(`phase1-T4-maximal-law.md` §2); the entire difference between the two is the\nlag-H cancellation between ρ(x) and ρ(x+H), and the fact V1 (`g | H` kills a\npair's contribution) says that is where all the cancellation lives. RML prices\nthe certificate with that cancellation FORFEITED, which is exactly why it is\nlegal and why it can only reach exponents, never TPC.\n\n**The owning-convention statement (per `SEARCH-CONVENTIONS.md` §1).** For the\ntwo-class rough set {n : n(n+2) coprime to P(z)} — OEIS A144311's object, the\n\"bounded number of residue classes per prime\" convention of MO 88323 — RML is\na position-uniform sup-norm bound on the signed remainder of the linear-sieve\nupper/lower approximants composed over the divisor-pair lattice: the κ = 2\nanalogue of the error analysis that powers Iwaniec's Jacobsthal bound\n(g(q) ≪ log²q, exponent 2 at primorials, κ = 1; Iwaniec 1978, carried per\n`sift-limit-attack.md` §4.5's account of the 1971/1980 engine). The statement\nclass it would strengthen: **Jacobsthal-type upper bounds for integers\navoiding two residue classes per prime**, where the only standing exponent is\nβ₂ = 4.26645 (DHR 2008 sifting limit, composition `paper/beta2-note.md`),\nunimproved since 2008 and with NO published bound at any exponent for the\ntwo-class case. For a proof attempt, the estimate's own owning convention is\n**bilinear/trilinear forms with Kloosterman fractions** (§1 row: DFI 1997;\nBettin–Chandee 2015).\n\n---\n\n## 2. THE PRICING LEMMA, WITH THE RIDERS PAID\n\n**Lemma (schema).** Assume RML(α) with α < β₂. Set H(z) = ⌈(2C·z^α + 1)/M⌉.\nThen for every z ≥ z₀ and every position x, T(x) ≥ 1, hence every window of\nlength H contains a twin-admissible r, hence\n\n G₂(z#) ≤ H(z) ≪ z^α · ln²z ≪ z^{α+ε}.\n\nChain, graded: (i) T = HM + R_H is a PROVEN identity; (ii) |R_H(x)| =\n|ρ̃(x) − ρ̃(x+H)| ≤ 2 sup|ρ̃| — the mean cancels exactly in the difference, so\nthe centered sup is the right object; (iii) HM − 2 sup|ρ̃| ≥ 1 by the choice\nof H; (iv) the Brüdern–Fouvry pointwise inequality makes T(x) a lower bound\non the twin-admissible count in (x, x+H] [PROVEN, pilot-verified]; (v)\nM ≥ c(s)/ln²z for s = 3.0 > 1+√e [standard linear-sieve asymptotics; measured\nM·ln²z = 0.34–0.37 flat across z = 13..47, consistent]. The only unproven\ningredient is the law itself.\n\n**Rider 1 — T ≥ 0 vs T ≥ 1, paid explicitly.** The fixed point of\n`f(H) = ⌈sup|R_H|/M⌉` certifies T ≥ 0, which certifies nothing\n(`theta-selfconsistent.md` reading 8; the operative window is the last fixed\npoint plus one). The lemma therefore uses the strict form throughout: the +1\nin (2S+1)/M is the certificate-side transplant of `--strict`'s\n`g(H) = ⌈(sup|R_H|+1)/M⌉`. Its price is Δneed = 1/M = O(ln²z), measured\n17.9 → 43.2 positions across z = 13..47, i.e. d_strict = 0.04467 → 0.00079 in\nexponent: carried, and never the difference between verdicts.\n\n**Rider 2 — the floor.** Positivity is observed only from nP(z) onward, so no\ncorrect law delivers a window below nP: the deliverable exponent band at z is\n[th(nP), β₂). With the corrected instrument's values the floor is 1.5963,\n1.7070, 1.7960, 1.7710, 1.7718, 1.8569 at z = 13..31 (exact) and ≥ 2.0106 at\nz = 47 (prefix bound). **The (31, 47] crossing is the floor passing the z²\nzone budget: it removes the TPC-grade deliverable and consumes 0.5% of the\nband against β₂** — at z = 47 the band is still 2.256 exponents wide. The\ncrossing is a fact about what a PERFECT law could deliver, not about whether\na weak law exists; for this item it prices the bottom of the band, and the\ntop (β₂) is untouched.\n\n---\n\n## 3. THE SUFFICIENCY CURVE\n\nFrom the producer's S3 (exponents base z; d_strict included above):\n\n```\n z theta_G lambda_max factor z^lam alpha_max th(nP) floor\n 13 1.94696 2.31949 383.5 2.8712 1.5963\n 17 2.01948 2.24697 581.8 2.9425 1.7070\n 19 2.15885 2.10760 495.6 2.9344 1.7960\n 23 2.18269 2.08376 687.9 2.9685 1.7710\n 29 2.15889 2.10756 1208.1 3.0369 1.7718\n 31 2.25232 2.01413 1008.8 3.0361 1.8569\n 37 2.34513 1.92132 1030.4 3.0848 —\n 41 2.38567 1.88078 1079.7 3.0995 —\n 43 2.43473 1.83172 981.9 3.0954 —\n 47 2.48199 1.78446 963.4 3.1085 ≥2.0106\n```\n\nReading the columns: `theta_G` is what the Gaussian law (F4 at λ = 0) buys —\nTODO 0's \"measured at 2.43 at z = 43 and rising\", reproduced here (2.43473).\n`lambda_max = β₂ − theta_G` is the curve itself: **how much weaker than\nGaussian, in z-exponent, a provable law may be at that z and still beat β₂.**\n`alpha_max = log_z((z^{β₂}M − 1)/2)` is the finite-z absolute ceiling on\nsup|ρ̃| with constant 1 (asymptotically it tends to β₂; at measured z it reads\n~3.1 because the ln²z and the constants bite at small z). The weakest measured\npoint is z = 47: λ_max = 1.78446, a factor of 963.\n\n**Calibration of how much room that is.** The trivial ℓ¹ law\nsup|ρ̃| ≤ (n+M)/2 — no cancellation at all, pure triangle inequality —\nalready suffices at every measured z ≤ 47 (need_triv/z^{β₂} = 0.27 rising to\n0.9323 at 47) and dies at z ~ 64 on the window fit\n(d ln need_triv/d ln z = 5.1178 ± 0.2161). This is the Bonferroni x ≤ 227\nphenomenon at a new instrument: at small z the β₂ comparison flatters any\nbound, so finite-z sufficiency is a calibration and NOT a result. What it\ncalibrates: near the measured range the asymptotic ask over\nabsolute-value accounting is under one unit of exponent, while the Gaussian\ntruth (if it persists) leaves ~2.3 units of measured cancellation — the\ntheorem's whole difficulty is asymptotic, and the window between \"trivially\ntrue at finite z\" and \"provable for all z\" is where every route here has\ndied.\n\n**Between Gaussian and sufficient, the measured margin.** The new S1\nmeasurement puts the truth at C_true = 0.60–0.80 of the Gaussian form at\nz = 13..29 (full periods; the walker verified four ways). So at measured z the\ndistance from TRUTH to the sufficiency ceiling is the full λ_max plus the\n20–40% Gaussian margin; a law loose by three orders of magnitude at z = 47\nstill finishes. Nothing about that margin is asymptotic; it is stated to\nshow the target's slack is enormous where it can be seen at all.\n\n---\n\n## 4. WHAT IS NEW IN THE MEASUREMENT COLUMN\n\nsup|ρ̃| itself had never been measured (every prior supremum in the corpus is\nof R_H at some window). Exact over complete periods:\n\n```\n z W sup|rho~| sqrt sqrt(2lnW) C_true\n 13 2,310 2.62013 1.04645 3.9357 0.6362\n 17 30,030 4.33665 1.58005 4.5409 0.6044\n 19 510,510 9.15247 2.22539 5.1270 0.8022\n 23 9,699,690 12.10617 2.82533 5.6723 0.7554\n 29 223,092,870 17.90249 3.68727 6.2005 0.7830\n```\n\nThe Gaussian ρ maximal law holds at every measured level, margin 20–40%, no\ntrend a five-point column can call. Contrast: the SHARP law (same template on\nR_H at the operative window) is violated at z = 19 by 0.6% (T4 §4). The legal\nlaw is, so far, true; the TPC-implying law is, already, false as literally\nstated. That asymmetry is consistent with T4's diagnosis that the sharp form\nis \"the wall wearing a different hat\" while the ρ form genuinely forfeits the\nlag cancellation.\n\nz = 31 was not walked (W = 6.47e9, ~500 s by the theta-selfconsistent rate;\nthe machine stays light tonight). It is the next exact point.\n\n---\n\n## 5. THE LITERATURE POSITION\n\nPosition assembled from the standing registries (`SEARCH-CONVENTIONS.md`\n§§1, 3, 4; `PRIOR-ART.md`; `REFUTED.md`). **No new literature search was run\ntonight**; every row below carries its own prior verification date, and\nabsences are quoted with where the corpus looked.\n\n**The object has no published bound.** No published upper bound on the\ntwo-class Jacobsthal function exists at ANY exponent, conditional or not —\nsearched in the owning convention (A144311's wording; MO 88323's \"bounded\nnumber of residue classes per prime\"), 2026-08-18, re-verified independently\nthe same day (`SEARCH-CONVENTIONS.md` §3; `sift-limit-attack.md` §5). β₂\nitself is unimproved since DHR 2008 and everything published after is worse\nat κ = 2 (Blight 4.45, Franze 4.516, Opera de Cribro 4.83; §4 table). So any\nRML(α<β₂) proof would not strengthen a published two-class statement — it\nwould create the first one below the sifting-limit default.\n\n**The one-class lineage the law parallels.** Iwaniec's g(q) ≪ log²q — at\nprimorials, exponent 2 — is the κ = 1 instance of exactly this pipeline:\nlinear sieve plus remainder control uniform in position. At κ = 1 the trivial\nremainder accounting already suffices for exponent 2+ε (level below window);\nat κ = 2 the coupled accounting gives 5.2974 > β₂, which is WHY a genuine\ncancellation law is the ask. The MO 37679/52890 route (zeb 2011) derives\nj(x#) ≪ x^{4.032} from a sieve error exponent at dimension one — the shape of\nour pricing lemma, one dimension down; per §5's discipline its 4.032 is never\ntabled beside 4.2665.\n\n**The strongest print for the operative remainder, priced.** The remainder\nreduces exactly (BigInt-verified, 2655 pairs) to trilinear forms with\nKloosterman fractions; the published estimates read γ = 1.009638 (DFI 1997,\nworse than trivial at our level split) and γ = 0.970624 (Bettin–Chandee\n2015) against a needed 0.824975 — in θ_total currency 1.0303 delivered\nagainst 1.2090 needed, returning exponent **5.0907, i.e. 0.824 of exponent\nABOVE the sufficiency ceiling** (`sift-limit-attack.md` §4.5;\n`history/staging/attack-sqrt-cancellation.md`). The position-uniformity gap is\nstructural on top of that (their Remark 1's charge is O(1) only for\nx ≪ H^{1.21} while our x reaches exp(H^{0.2344})). Caveat, stated rather than\nhidden: that pricing was computed at the R_H/Lemma-V decoupled configuration;\nthe ρ-potential configuration was not separately priced tonight (NOT\nREACHED), and nothing suggests the sup over a full period is cheaper — it is\nthe same lattice with a longer position range.\n\n**Verdict for (c): the sufficiency curve's weakest point is ABOVE published\nstrength everywhere.** Published instruments deliver 5.09+ where the ceiling\nis 4.26645; published two-class bounds do not exist; the smooth-modulus and\nalgebraic-geometry branches are CLOSED on hypotheses (`REFUTED.md` rows,\n2026-08-20); the field's special-level constructions certify nothing on the\nsmallness side (`special-levels-recon.md`). **No headline: nothing in print\nimplies even the weakest sufficient form.** The quantified distance from the\nprint frontier to the ceiling is 0.1787 of θ_total (1.2090 − 1.0303), the\nsame frontier gap already recorded for Lemma V — the ρ reframing does not\ncheapen the estimate; what it changes is the LEGALITY (not TPC-implying) and\nthe win condition (any α < 4.26645, rather than the zone budget).\n\n**HELD, one adversarial pass wanted before this moves anywhere live:** at the\nsufficiency ceiling the working point is s/u = 3/4.26645 = 0.703, inside\nLemma V's stated range s ≥ (0.63+δ)u — and every delivered exponent in\n(3, β₂) stays inside it. TODO 0's \"moot and moving further away, s/u =\n1.62–1.72\" is a statement about TPC-grade windows and is correct there; for\nthe β₂ target the working point RE-ENTERS the range Lemma V was posed for.\nIf the pass confirms, the Lemma V sub-question is un-mooted FOR THIS ITEM\n(in its weakest F2 form, at u near β₂ only), and the κ = 1 precedent\n(Iwaniec 1980) plus the Kloosterman toolbox are pointed at the right ask for\nthe first time: a 0.18-of-θ_total improvement, not a factor-2 miracle.\n\n---\n\n## 6. READINGS\n\n1. **STATED.** The ρ maximal law in five forms (F0–F4), quantifiers explicit,\n in corpus vocabulary and the owning convention; the pricing lemma\n RML(α<β₂) ⟹ G₂(z#) ≪ z^α ln²z with only the law unproven; both riders\n (T ≥ 1 accounting; the nP floor) paid explicitly. §§1–2.\n\n2. **MEASURED.** The sufficiency curve λ_max(z) = β₂ − θ_G(z) runs 2.31949 →\n 1.78446 across z = 13..47; weakest point z = 47, allowed looseness factor\n 963 over the Gaussian form. The strict accounting costs 0.04467 → 0.00079\n in exponent and never a verdict. Producer S3.\n\n3. **MEASURED, new.** The Gaussian ρ law is true at every exactly measurable\n level: C_true = 0.6362, 0.6044, 0.8022, 0.7554, 0.7830 at z = 13..29 over\n complete periods, walker verified against brute force, closed-form\n variance (1.2e-8 %), period mean, and period closure. First direct sup|ρ|\n in the corpus. Producer S1.\n\n4. **MEASURED, finite-z calibration only.** The trivial ℓ¹ law already\n suffices at every z ≤ 47 (ratio 0.27 → 0.9323) and dies at z ~ 64 on the\n window fit (slope 5.1178 ± 0.2161, estimator calibrated in-pass). The\n Bonferroni-x≤227 phenomenon; explicitly NOT a route and NOT a headline.\n Producer S4.\n\n5. **MEASURED / CITED.** The floor: th(nP) = 1.5963..1.8569 exact at\n z = 13..31, ≥ 2.0106 at z = 47. The (31, 47] crossing kills the exponent-2\n deliverable and consumes 0.5% of the room to β₂, leaving a 2.256-wide\n band. The crossing prices the floor, not the ceiling. Producer S5.\n\n6. **MEASURED, two models, NEITHER CALLED.** Where the room closes: linear\n drift (slope 0.3909 ± 0.0363; nine-point control reproduces T4's 0.3746)\n says z ~ 5·10³ [2.4·10³, 1.2·10⁴]; the ⟨ρ²⟩ power model (slope 2.2355 ±\n 0.1418 here, 2.4625 ± 0.2184 on theta-ladder §3's window) says θ_∞ =\n 2.74 ± 0.14 and the room never closes. Ten points over one octave decide\n nothing between them. Producer S6.\n\n7. **INFERRED from the standing registries (no new search).** Nothing\n published implies any sufficient form: best print delivers exponent\n 5.0907 against the 4.26645 ceiling (gap 0.1787 of θ_total), and no\n two-class bound exists in print at any exponent. The sufficiency curve\n sits above published strength at every point — the theorem is open in the\n honest direction, with the truth (measured) two exponents below the\n ceiling. §5.\n\n8. **HELD (one adversarial pass wanted).** The β₂-target working point\n s/u = 0.703 is inside Lemma V's stated range, for every delivered\n exponent in (3, β₂): the \"Lemma V is moot\" verdict is TPC-specific and\n does not transfer to this item. §5.\n\n## 7. NOT REACHED\n\n- **sup|ρ̃| at z = 31** (W = 6.47e9, ~500 s): the next exact point of the\n C_true column, deferred under tonight's compute ceiling.\n- **The conditional row at z = 47 recomputed** rather than cited (~2800 s\n O(N²)); the cited row is custody-bound and its cheap fields verify, but a\n full local reproduction was not bought.\n- **An analytic upper bound on ⟨ρ²⟩(z)** from its own closed form (a proven\n z^c would turn the F4 curve into an unconditional F3 statement and pin\n θ_∞); not attempted, and it is the natural next lemma-shaped object.\n- **Pricing the Kloosterman import at the ρ-potential configuration**\n (position range e^{θ(z)} instead of the Lemma-V window); the recorded\n pricing is at the R_H configuration and was carried over with that caveat\n stated.\n- **The s-freedom.** Everything here is s = 3.0. The crude column is nearly\n s-insensitive (`theta-ladder.md` §5a) but the sufficiency curve's s\n dependence — in particular whether some s < 3 lowers θ_G at large z — was\n not measured.\n- **The adversarial pass on the HELD s/u finding** (reading 8).\n\n## 8. GATE\n\n`node research/qc.js` after the producer and this report were written:\n**TOTAL 0** — no check names `research/rho-maxlaw-01-sufficiency.js` or this\nfile. The producer's tail verifies: `qc/embed.js --check` reports code-sha256\nmatches, out-sha256 matches, body bit-honest. (The standing advisories —\nembed-backlog, sourcing-backlog — are pre-existing and not part of TOTAL.)\n", "research/history/staging/xchan-at29.md": "# The blind @29 test of the joint-deficit closed form, and the segmented census that made it possible\n\n\n\n*(2026-08-19. TODO item X's `~3.8` constant had a zero-parameter closed-form\ncandidate with blind values pre-computed at two levels that did not exist, and a\ncensus instrument that stopped at @23 for want of 162 GB. This file builds the\nsegmented instrument, validates it against the in-memory one at five levels to\nthe unit, and spends both blind tests. Pre-registration, written and committed\nalone before any producer script existed:\n[xchan-at29-prereg.md](xchan-at29-prereg.md). Producer, formally embedded:\n`research/xchan-at29-01-segmented.js`. Every figure below is from that file's\nOUTPUT block, or is one-line arithmetic on figures in it, shown with its inputs.\nCalibration is marked on every claim: PROVEN, MEASURED, DERIVED, REFUTED.)*\n\n## Verdict\n\n**The closed form `1 − J = 4S₂` is the only one of the three pre-registered laws\nleft standing, and it is not exact.** At @29 it scores a clean HIT\n(`z = −0.90`); at @31, where the count is 1.65 billion and `σ_J` falls to\n`0.000024`, it is 4.93σ low. The two rivals the pre-registration fixed before the\nrun die at @31 at 6.35σ and 20.81σ. What the two blind levels agree on is a\n**stable relative offset of about half a percent**, `d = −0.41%` at @29 and\n`−0.48%` at @31, and the corresponding absolute residuals\n`0.028943 − 0.028823 = 0.000120` and `0.024784 − 0.024666 = 0.000118` are the\nsame number to 2%.\n\n**And the memory ceiling is gone.** The census now costs **62.0 MB of arrays at\nevery level**, and the level enters only the clock: @23 in **0.56 s** against the\nin-memory instrument's 169.89 s, @29 in **15.07 s**, @31 in **596.78 s**. The\nNamed gap \"@23 is the in-memory instrument's last level (@29 needs 162 GB)\" is\nclosed by construction, not by buying memory.\n\nFive things the run settles that the record could not:\n\n1. `J` is now measured at **seven levels** and is monotone from @17 through @31:\n 0.902452, 0.959855, 0.965932, **0.971177**, **0.975334**. Both new levels sit\n in `natal-cap-39`'s pre-registered DECAYING FASTER THAN FORCED band, so\n `xchannel-triples.md` R1's verdict rests on five points rather than three.\n2. **P1 is directly tested two and three levels beyond where it had ever been\n tested.** Aligned super-`W` obs is exactly 0 at @29 and at @31.\n3. The `1/lnW` reading of the decay, which was indistinguishable from the closed\n form at @23 and 0.4σ from it at @29, is **REFUTED at 20.81σ at @31**.\n4. The \"asymptote is exactly 4\" reading gains nothing. The measured ratio\n `(1−J)/F` runs **7.0401, 3.6905, 3.8630, 3.8240, 3.8427** over @17..@31: it\n does not rise, while the closed form's own ratio rises monotonically\n 3.7245 → 3.8612 over the same span.\n5. The candidate's residual, invisible at every level that scored it, is now\n **measured**: one detection at 4.93σ (@31) and one consistent non-detection\n at 0.90σ (@29).\n\n---\n\n## 1. What the in-memory instrument measures, and why @29 cost 162 GB\n\n`research/natal-cap-39-triple-census.js` defines, at level `x` with\n`W = ∏_{p≤x} p` and natal set `N = {r ∈ [0,W) : r ≡ 11,17 (mod 30), p ∤ r,\np ∤ r+2 for 7 ≤ p ≤ x}` and scour primes `qs = {q : x < q, q² ≤ W}`:\n\n> `J(x) = (Σ_{∏Q > W, mixed} c(Q,ε)) / (6·N̄·Σ_{∏Q > W} 1/∏Q)`,\n> `c(Q,ε) = #{r ∈ N : r ≡ −ε_q (mod q) ∀q ∈ Q}`, `ε ∈ {0,2}^Q`,\n\nwith `cls = #{q : ε_q = 0}`, `cls ∈ {0,3}` ALIGNED and `cls ∈ {1,2}` MIXED. It\nreaches the super-`W` part by the bridge\n`mixed super-W obs = Σ_{r∈N}[C(a,2)b + a·C(b,2)] − (sub-W mixed obs)`, with\n`a(r) = #{q | r}` and `b(r) = #{q | r+2}` carried as two `Uint16Array(W)` beside\na `Uint8Array(W)` natal mask. That is **5 bytes per slot of `[0,W)`**, and\n`natal-cap-35`'s version of the same pass is 25. At @29,\n`W = 6,469,693,230`, so 32 GB and 162 GB respectively — the number\n`xchannel-triples.md` §7 records, and the reason the series stopped.\n\nThe bridge's second term is the other half of the price: the sub-`W` census\nenumerates 4,517,592 triples at @23 and scans `A` at stride `∏Q` across the\nwhole of `[0,W)`, which is random access over the resident array by\nconstruction.\n\n## 2. The segmented instrument, and its two levers\n\n`research/xchan-at29-01-segmented.js`. Two changes, and neither approximates\nanything.\n\n**(L1) Exchange the order of summation.** Summing over `r` on the outside\ninstead of over triples:\n\n> `mixed super-W obs = Σ_{r∈N} #{(q₁ W}`\n> ` + Σ_{r∈N} #{(q₁ ∈ D_a(r), q₂ W}`\n\nwith `D_a(r) = {q ∈ qs : q | r}` and `D_b(r) = {q ∈ qs : q | r+2}`. Every `r`\ncontributes through its own two divisor lists and nothing else, so the census is\na sum of a bounded function of the local factorisation over a sieved interval —\nexactly the shape the corpus's segmented-sieve lever handles, and the same shape\n`research/import-scanstat-04-score.js` streams over these same 6.47e9 positions\nfor T₂₉ with a depth-64 ring buffer and no materialised gap word.\n**[MEASURED] The lists are short**: `max |D_a| = 5, max |D_b| = 6` at @29 and\n`6, 6` at @31, against the code's capacity of 7, which it checks.\n\n**(L2) Sieve the compressed slot line.** Every `r ∈ N` is `≡ 11` or `17`\n(mod 30), so only `W/15` of the line can matter. Indexing slots by `s = 2k + t`\nwith `r = 30k + (t ? 17 : 11)`, a scour prime `q` marks one class on one side at\n`r ≡ CRT(class, side) (mod 30q)` — every `2q`-th slot. The marking cost falls by\nthe full factor 15. At @29 that is **431,312,882 slots and 994,345,351 scour\nmarks** where the uncompressed line would need 1.49e10.\n\n**Exactness at the one inequality.** The kernel compares `q₁q₂q₃` to `W`. At @31\nthat product reaches 9.0e16 and is not exact in a double, so it is never formed:\n`U[m] = ⌊W/q_m⌋` is precomputed with an integer correction (the division is not\ntrusted), and the test is `q₁q₂q₃ > W ⟺ q₁q₂ > U[m]`, valid because every scour\nprime exceeds `x` so `q_m ∤ W` and `W/q_m` is never an integer. `q₁q₂ ≤ 2.0e11`\nis exact.\n\n**The denominator is computed as a tail, never as a difference.**\n`Σ_{∏Q>W} 1/∏Q` is accumulated directly over the super-`W` triples by a monotone\npointer on the third prime, so no cancellation enters it. Its consistency with\nthe other route is printed anyway: `miss + sub = e₃` agrees with the elementary\nsymmetric function to eight decimals at every level (0.25113952 at @29,\n0.33095269 at @31), and the triple counts sum to `C(K,3)` exactly\n(80,993,070,011 at @29; 8,812,286,043,484 at @31).\n\n**Memory.** `A` (1 byte) + `a` (1) + `b` (1) + two `Uint16` divisor lists of\ncapacity 7 (28) = **31 bytes per slot on 2²¹ slots = 62.0 MB, at every level**.\nThe level enters only the segment count: 206 at @29, 6,376 at @31.\n\n## 3. The validation gate — the instrument reproduces the in-memory one exactly\n\nPrereg §5 fixed this before the producer was written: no @29 number is reported\nunless the new instrument reproduces `natal-cap-39`'s embedded OUTPUT to the\nunit at @11, @13, @17, @19 and @23. Fourteen integers and three rounded reals per\nlevel, 85 comparisons in all.\n\n| level | `N̄` | `B₃` split (3,0)/(2,1)/(1,2)/(0,3) | sub-`W` triples | sub-`W` mixed | **mixed super-`W`** | (2,1)/(1,2) | `Σ_{∏Q>W}1/∏Q` | `J` |\n|---|---|---|---|---|---|---|---|---|\n| @11 | 90 | 0/2/0/0 | 0 | 0 | **2** | 2/0 | 0.006909 | 0.536070 |\n| @13 | 990 | 12/89/70/11 | 71 | 16 | **143** | 77/66 | 0.023518 | 1.023632 |\n| @17 | 14850 | 848/2729/2713/847 | 5246 | 2243 | **3199** | 1622/1577 | 0.039784 | 0.902452 |\n| @19 | 252450 | 27377/85713/85449/27391 | 162982 | 97097 | **74065** | 37139/36926 | 0.050943 | 0.959855 |\n| @23 | 5301450 | 879225/2779604/2779107/879385 | 4517592 | 3751046 | **1807665** | 904409/903256 | 0.058834 | 0.965932 |\n\n**[MEASURED] GATE: 5 reference levels checked, 0 mismatches.** Every integer,\nand `Σ_{∏Q>W}1/∏Q`, the CRT denominator and `J` at the record's own printed\nprecision. The two instruments share no data structure: one enumerates triples\nand scans a resident `W`-wide mask, the other sieves a compressed slot line in\n62 MB segments and never enumerates a triple in the numerator at all.\n\nTwo further identities hold at every level including the two new ones:\n`sub + sup = B₃` class by class (INTERNAL IDENTITY PASS), and the counted `N̄`\nequals `2∏_{7≤p≤x}(p−2)` (143,139,150 at @29; 4,151,035,350 at @31).\n\n## 4. The price, and what it now costs to go up a level\n\n| level | slots | scour marks | segments | census | level total |\n|---|---|---|---|---|---|\n| @23 | 14,872,858 | 30,527,907 | 8 | 0.55 s | **0.56 s** (natal-cap-39: 169.89 s) |\n| @29 | 431,312,882 | 994,345,351 | 206 | 13.02 s | **15.07 s** |\n| @31 | 13,370,699,342 | 33,742,377,278 | 6,376 | 544.68 s | **596.78 s** |\n\nWhole file, seven levels: **612.6 s**. The difference between the census column\nand the level column is the denominator's `K²` pass, 2.05 s at @29 and 52.10 s\nat @31.\n\n**[MEASURED] The factor at @23 is 303.** The in-memory census needed 169.89 s and\n1.203 GB there (`xchannel-triples.md` §3); this one needs 0.56 s and 62.0 MB, and\ngets the same integers. The @31 leg cost 596.78 s, which is why the prereg's\nconditional @31 clause was executed rather than deferred.\n\n**Timings are the only thing here that moves.** The file was run twice end to end\non a machine shared with four concurrent siblings, and the two runs have the\n**same normalised output hash** (`out-sha256 0c5f57087dd86f49…`): every integer,\nevery ratio and every verdict is identical, and only the wall clock differs.\n\n## 5. @29 and @31, scored against the pre-registration\n\nThe predictions are `import-stein.md` §3.2's, adopted verbatim in prereg §2 and\nhard-coded in the producer. **[MEASURED] Custody first:** `4S₂` recomputed here\nfrom the level's own prime list is **0.028943** at @29 and **0.024784** at @31,\nagreeing with the record's blind values to the printed digit. The prime lists are\n`K = 7863` and `K = 37534`, which are the record's own counts, running 31..80429\nand 37..447829 inside the intervals (29, 80429] and (31, 447829] it names, and\n`F` likewise reproduces at 0.7537% and 0.6419%.\n\n| level | mixed super-`W` obs | CRT | `J` | `1 − J` | `σ_J` | predicted `1−J` | **z** | **d** |\n|---|---|---|---|---|---|---|---|---|\n| @19 | 74,065 | 77,162.70 | 0.959855 | 0.040145 | 0.003527 | 0.041122 | −0.28 | −2.38% |\n| @23 | 1,807,665 | 1,871,421.20 | 0.965932 | 0.034068 | 0.000718 | 0.033678 | +0.54 | +1.16% |\n| **@29** | **53,660,192** | **55,252,747.16** | **0.971177** | **0.028823** | **0.000133** | **0.028943** | **−0.90** | **−0.41%** |\n| **@31** | **1,653,241,687** | **1,695,051,393.52** | **0.975334** | **0.024666** | **0.000024** | **0.024784** | **−4.93** | **−0.48%** |\n\n(The @19 and @23 rows are this instrument's own reproduction of the record; the\nrecord scores them −0.29 and +0.60 from its rounded `1−J`.)\n\n**PRE-REGISTERED VERDICTS, printed by the producer's own classifier.**\n\n- **@29: TEST 1 `z = −0.90` → HIT. TEST 2 `d = −0.41%` → TIGHT.** Combined\n verdict, prereg §3: **HIT**. The J-band clause does not fire.\n- **@31: TEST 1 `z = −4.93` → MARGINAL. TEST 2 `d = −0.48%` → TIGHT.** Combined\n verdict, prereg §3: **SURVIVES WITH A MEASURED OFFSET — right to `d`, not\n right.** The J-band clause does not fire.\n\n**[MEASURED] R1 — the offset is the finding, and it is one detection, not two.**\n`d` is −0.41% at @29 and −0.48% at @31, and the absolute residuals\n`0.028943 − 0.028823 = 0.000120` and `0.024784 − 0.024666 = 0.000118` agree to\n2%. But `σ_J` is 0.000133 at @29, so **@29 does not detect the offset** — it is\nconsistent with it at 0.90σ and equally consistent with zero. The detection is\n@31's alone, at 4.93σ. The honest statement is: *one level measures a residual\nof 1.18e−4, a second level is consistent with the same residual and cannot\nresolve it, and the two levels that fitted the closed form had error bars\n0.003527 and 0.000718, respectively 30 and 6 times too wide to have seen it.*\nNo closed form is proposed for the residual here; two points, one of them a\nnon-detection, do not license one.\n\n**[REFUTED] R2 — the two rival laws, fixed in prereg §4 before the run.**\n\n| law | @29 prediction | z at @29 | @31 prediction | z at @31 |\n|---|---|---|---|---|\n| N1, the candidate `4S₂` | 0.028943 | **−0.90** | 0.024784 | **−4.93** |\n| N2, constant multiple of `F` fixed at @23 | 0.029145 | −2.43 | 0.024818 | −6.35 |\n| N3, `1 − J ∝ 1/lnW` anchored at @23 | 0.028990 | −1.26 | 0.025165 | −20.81 |\n\nPrereg §4 registered in advance that @29 could not separate these — N3 sits 0.4σ\nfrom the candidate there — and that @31 could. It did. **N3 is refuted at 20.81σ\nand N2 at 6.35σ, and the candidate is nearer than either at both blind levels.**\nThe closed form is the last law standing, and it is also refuted as exact.\n\n**[MEASURED] R3 — the ratio does not rise, and \"asymptote exactly 4\" gets no\nsupport.** The measured `(1−J)/F`, on the producer's own `F`:\n\n| x | 17 | 19 | 23 | **29** | **31** |\n|---|---|---|---|---|---|\n| measured `(1−J)/F` | 7.0401 | 3.6905 | 3.8630 | **3.8240** | **3.8427** |\n| `4S₂/F` | 3.7245 | 3.7804 | 3.8189 | **3.8399** | **3.8612** |\n\nThe closed form's ratio rises monotonically toward 4 by construction; the\nmeasured one falls from @23 to @29 and rises again to @31, inside a spread of\n0.04. **None of that shape is resolved**: the @23 ratio's own bar is ±0.081\n(`xchannel-triples.md` R3), which swallows the whole spread. What IS resolved is\nthe vertical offset at the two sharp levels, where `σ_J/F` is\n0.000133/0.007537 = 0.018 and 0.000024/0.006419 = 0.0037: the measured ratio sits\n0.0159 below the formula's at @29 and 0.0185 below it at @31, which is the\nprimary test's 0.90σ and 4.93σ read in another column. So the \"asymptote\nexactly 4\" clause in `import-stein.md` §3.2 is a property of the formula and not\na measured trend, and the one thing the measurement does say about it is that\nthe formula runs high.\n\n**[MEASURED] R4 — `J` itself, seven levels, monotone from @17.** 0.536070,\n1.023632, 0.902452, 0.959855, 0.965932, **0.971177**, **0.975334**, with\nincrements from @17 of +0.057403, +0.006077, **+0.005245**, **+0.004157**. Both\nnew levels classify DECAYING FASTER THAN FORCED under `natal-cap-39`'s\npre-registered bands (0.945 < J ≤ 0.995), and the monotonicity clause holds. The\ndeficit is still large in its own units: at @31 the shortfall is\n1,695,051,393.52 − 1,653,241,687 = **41,809,706.52 coincidences, more than\n1,000 σ_J from zero**. Decaying is not absent, and two more levels do not make\nit absent.\n\n**[PROVEN, now MEASURED twice more] R5 — P1 holds at @29 and @31.** Aligned\nsuper-`W` obs is **exactly 0** at both, by direct enumeration of the divisor\nlists rather than by the bridge subtraction that produced the @19 and @23 zeros.\n\n**[MEASURED] R6 — the two orientations stay symmetric.** (2,1) and (1,2) split\n26,827,582 / 26,832,610 at @29 (0.019% apart) and 826,645,240 / 826,596,447 at\n@31 (0.006% apart), against 27,626,373.58 and 847,525,696.76 each. The deficit\nis not carried by one orientation at either new level.\n\n## 6. What this does not show\n\n- **No derivation.** `import-stein.md` §3.3 killed the Chen–Stein route\n (`b₃/(b₁+b₂) = 73.4` at @23) and §3.4 refuted the super-`W` decomposition\n route. Nothing here revives either. The closed form is still a formula that\n fits, now known to be accurate to about half a percent and no better.\n- **The residual is not identified.** One 4.93σ detection at one level. Any\n closed form fitted to it now would be fitted to a single number.\n- **N2 is not fully separated from N1 by structure, only by value.** A constant\n multiple with the constant re-fitted at @29 or @31 rather than at @23 would\n track the measurements as well as `4S₂` does; what @31 refutes is the specific\n constant the prereg fixed at @23.\n- **The `σ_J` error model is inherited, not derived.** These counts are exact\n arithmetic; `√obs/CRT` is the record's own Poisson proxy for their\n fluctuation, and every z here depends on it. A different fluctuation model\n moves the 4.93σ.\n- **`j = 4` is still untouched**, and the steelman is stated at `i + j ≥ 3`.\n Its missing mass at @23 is 0.040035, and the same instrument would reach it.\n- **No σ_X, no rotation ensemble, no `m`-decomposition.** This file computes the\n joint census and nothing else; `natal-cap-39`'s other columns are not\n reproduced here and are not claimed.\n- **No live document carries any of this.** `TODO.md` item X, `paper/anchored-note.md`,\n `research/moire-theorems.md`, `import-stein.md` and `xchannel-triples.md` were\n not touched. §7 proposes the item X edit; it is not applied.\n\n## 7. Proposed update to TODO item X (report only, not applied)\n\nThree edits, and one Named gap to close.\n\n**(a) The closed-form clause.** Replace *\"the `~3.8` constant now has a\nZERO-PARAMETER closed-form candidate ... landing at −0.29σ (@19) and +0.60σ\n(@23) with asymptote exactly 4 ..., and blind values pre-computed at @29 ... and\n@31 .... so the derivation stays the item's open analytic target, now with a\ntarget formula and two blind tests waiting\"* with:\n\n> **BOTH BLIND TESTS SPENT, 2026-08-19** (`research/history/staging/xchan-at29.md`,\n> pre-registered in `xchan-at29-prereg.md` and committed alone before the\n> producer existed). `1 − J = 4S₂` measures **0.028823 at @29 against the\n> predicted 0.028943 (z = −0.90, HIT)** and **0.024666 at @31 against 0.024784\n> (z = −4.93, d = −0.48%)**. It is the **only survivor of the three laws the\n> prereg fixed**: a constant multiple of the forced scale fixed at @23 dies at\n> 6.35σ and `1 − J ∝ 1/lnW` at 20.81σ, both at @31. It is also **refuted as\n> exact** — the two blind levels share a relative offset of −0.41% and −0.48%,\n> absolute residuals 1.20e−4 and 1.18e−4, detected at 4.93σ at @31 and only\n> consistent at 0.90σ at @29. **\"Asymptote exactly 4\" is untested and gains\n> nothing**: the measured ratio runs 7.0401, 3.6905, 3.8630, 3.8240, 3.8427 over\n> @17..@31 and does not rise, while the formula's rises 3.7245 → 3.8612. The\n> derivation remains the item's open analytic target and now has to produce the\n> half-percent offset as well as the leading term.\n\n**(b) The `J` series.** *\"the mixed super-W joint deficit runs J = 0.9025,\n0.9599, 0.9659 at @17/@19/@23\"* becomes *\"...0.9025, 0.9599, 0.9659, 0.9712,\n0.9753 at @17/@19/@23/@29/@31, monotone at all five, all three post-@17 levels\nin the DECAYING FASTER THAN FORCED band, and P1 (aligned super-W ≡ 0) now\ndirectly enumerated at @29 and @31 as well.\"*\n\n**(c) Named gaps.** Strike *\"@23 is the in-memory instrument's last level (@29\nneeds 162 GB)\"* and replace with:\n\n> The census is segmented (`research/xchan-at29-01-segmented.js`): **62.0 MB of\n> arrays at every level**, @23 in 0.56 s against the in-memory instrument's\n> 169.89 s and to the same integers, @29 in 15.07 s, @31 in 596.78 s. The memory\n> ceiling is gone and the clock is what is left: @29 → @31 cost a factor of 40\n> for a factor of 31 in `W`, so @37 is an overnight run rather than an\n> impossibility (extrapolated, not priced).\n\n**(d) The first move.** *\"decide whether δ's sign flip is finite-size or\nstructure before any segmented @29 rewrite\"* is spent: the rewrite exists and\nwas cheaper than the decision it was gated behind. The cheap next instrument\nnamed there, *\"j = 4's own census, missing mass 0.040 at @23\"*, is unchanged and\nis now reachable at @29 and @31 by the same code path.\n\n## 8. Files touched\n\n- `research/history/staging/xchan-at29-prereg.md`: new, committed **alone**\n before the producer existed (`27dc709`, \"prereg: xchan-at29\").\n- `research/xchan-at29-01-segmented.js`: new. Tail embedded with\n `node research/qc/embed.js --timeout 1800 research/xchan-at29-01-segmented.js`.\n **The file carries a PRESENTATION NOTE in its header and the reason is on the\n record inside it.** Each blind level was computed once before the canonical\n embed run, with these predictions and these bands already hard-coded and the\n pre-registration already committed, so neither figure was ever produced by an\n unclassified run. @29 was first computed on a revision differing only in two\n OUTPUT lines (the custody line, and a corrected scour-mark work count, which is\n printed and not used); @31 was first computed on the present revision, before\n the default level list gained 31. No band, rule, threshold, prediction or\n computed quantity changed at any point, and the embedded block reproduces both\n first runs digit for digit. The file was embedded twice: the first embed's\n header quoted the @31 probe's own census time, which the canonical run does not\n reproduce because the machine is shared with four concurrent siblings, so the\n figure was replaced by a pointer to the block's own price and the file re-ran.\n- `research/history/staging/xchan-at29.md`: this file.\n\nNo live document was edited.\n\n## 9. Gate\n\n**Before:** `node research/qc.js` TOTAL = **0**.\n\n**After:** the count **oscillates between 0 and 6** depending on the minute it\nis read, and every finding in it is `uncited-script`. `node research/qc.js\n--full` reports **117/117 checks passed** and fails the gate only on whatever\nthat TOTAL happens to be.\n\n**Reporting \"0 before, 0 after, unchanged\" would be true of a number and false\nof the fact.** The tree is shared with four concurrent siblings, as\n`xchannel-triples.md` §9 describes, and the finding set moved four times while\nthis ran: six (five siblings' scripts and this session's\n`research/xchan-at29-01-segmented.js`, waiting for the report that cites it),\nthen four (one sibling's `fdecay-deep` family), then zero, then four again as\nthat sibling's tree changed under the reading. **Since this file was written,\nnone of the findings has been this session's**: it cites its own producer, and\nthe `refs`, `quotes`, `crosslinks`, `scripts`, `transfers`, `calibration`,\n`absence`, `sourcing`, `embeds`, `provenance` and `search-convention` checks are\nall clean on both files this session wrote. `embeds` in particular is clean\nafter the numbered READINGS were appended below the OUTPUT banner, which is what\nconfirms the `code-sha256` binding was not disturbed by them.\n\n`node research/qc/embed.js --check --timeout 1800\nresearch/xchan-at29-01-segmented.js`: **code-sha256 matches, out-sha256\nmatches**, on a third full execution of the file.\n", "research/history/staging/xchannel-at23.md": "# The X-channel at @23: the fifth point of (X̄−X(0))/S̄\n\n\n\n*(2026-08-19. TODO item X's named first move, carried out. Every figure below\ncomes from `research/natal-cap-35-x-multiplicity.js` as re-embedded today, or\nfrom the two scratch instruments named in §2, which are not repo files.\nCalibration is marked on every claim: MEASURED, DERIVED, INFERRED, REFUTED.)*\n\n## Verdict\n\n**The fifth point is `(X̄−X(0))/S̄ = +0.2658` at @23. The monotone rise holds at\nfive levels, and the rise is larger than the four-point trend predicted.**\nThe price was 18.9 s and 5.69 GB for the @23 level, 94.6 s for the whole file,\nagainst a 4-hour ceiling: affordable by a factor of about 760.\n\nThree things the run settles that four points could not:\n\n1. **The deficit has migrated into m ≥ 3 and is now dominant there.** The\n m ≥ 3 share of the X-gap runs 10.5%, 23.5%, **81.1%** at @17, @19, @23. The\n file header's warning that the anchored deficit lives at m ≥ 3 was formed\n at @17, where the m = 2 cell was in fact carrying 89.5% of the gap. At @23\n the warning is true of the gap itself for the first time.\n2. **The m ≥ 3 mechanism candidate is REFUTED as separate structure.** At the\n anchor, `ω_s(r)` and `ω_s(r+2)` are independent to 0.05% on X. There is no\n room left for \"coincidences at multiples of products, dense at the origin\"\n as a distinct effect: the one-sided cofactor law already carries 84% of the\n m ≥ 3 gap, and part4 gives that law in closed form.\n3. **@23 is this instrument's last level, and the blocker is memory.** 25 bytes\n per slot × W. @29 would need 162 GB. A sixth point needs a rewrite, not a\n bigger budget.\n\n---\n\n## 1. Assignment\n\nExtend `(X̄−X(0))/S̄` from four levels to five. TODO item X names the edit\n(`for(const x of [11,13,17,19])`) and the discipline (\"Price the run, not the\nplumbing\"). Standing state before the run:\n\n| x | (X̄−X(0))/S̄ | X(0)/X̄ | S(0)/S̄ | β | S_CRT/S̄ |\n|---|---|---|---|---|---|\n| 11 | −0.2287 | 1.4541 | 1.1768 | 1.1458 | 1.0271 |\n| 13 | +0.0134 | 0.9908 | 0.9875 | 1.0089 | 0.9788 |\n| 17 | +0.1568 | 0.9482 | 0.8573 | 0.9549 | 0.8978 |\n| 19 | +0.2225 | 0.9558 | 0.7795 | 0.9261 | 0.8416 |\n\nAssumption A, in the channel it actually lives in: *the anchored overlap-credit\ndeficit stays below (1−ε)·S̄*, which is the first column staying below 1.\n\n## 2. Instrument\n\n`research/natal-cap-35-x-multiplicity.js`, one line changed:\n`for(const x of [11,13,17,19,23])`. Tail re-embedded with\n`node research/qc/embed.js research/natal-cap-35-x-multiplicity.js --streams\nboth --timeout 600`; `--check` reconfirms both hashes. The file's previous tail\nwas a real embed, not a legacy hand-paste: `qc.js embed-backlog` classed it\n`readings-not-traceable` (advisory, 5 of 73 figures, all prose arithmetic), not\n`hand-pasted-tail`.\n\nTwo scratch instruments, in the session scratchpad and deliberately not repo\nfiles, since the task's edit budget is one line:\n\n- `xchan-price-params.js` counts the work each leg does at each level, and\n `xchan-probe.js` times the two dominant kernels at @19 and @23 array sizes.\n Both are pricing tools; §3 is their output.\n- `xchan-m3-closedform.js` runs the anchored-independence test of §7. It\n rebuilds the level from the same definitions rather than importing them, so\n its S(0) and X(0) are an independent recomputation, and they agree with the\n embedded block at all five levels.\n\n**What the instrument computes at @23 and what it does not.** `part0`, the\nfull W-rotation custody sweep, is gated on cap-31 having published a reference\nfor the level, which it did only for x ∈ {11,13,17}. So @23, like @19, gets no\nsweep: no per-rotation variances, **no σ_X**, every `z(anch)` NaN, no CUSTODY\nline. `part5(·,3)`, the j = 3 coincidence census, is gated at x ≤ 17, so the\ndirect aligned-versus-mixed test of P1 is not extended either.\n\n## 3. The price, derived before the run\n\nThe only superlinear leg is the custody sweep, O(W·N̄·H). At @23 that is\n2.23e8 rotations × 5.30e6 slots and is dead by roughly fourteen orders. It does\nnot run, so the price is the u-form path, whose legs and their exact work\ncounts fall straight out of the code:\n\n| leg | what it counts | @19 | @23 | ratio |\n|---|---|---|---|---|\n| `level` | O(W) sieve + comb build | 9.70e6 | 2.23e8 | 23.0 |\n| `uform` | `W·H` strided increments | 1.73e7 | 4.58e8 | 26.4 |\n| `part3` | O(W) + subsets with ∏Q < W | 261,332 | 6,385,817 | 24.4 |\n| `part4` | `W·Σ(1/q)` + a sieve to W | 8.66e6 | 2.29e8 | 26.4 |\n| `part5` j=2 | `4·W·Σ_{pairs}1/(q₁q₂)` | 1.53e7 | 4.66e8 | 30.6 |\n\nW goes ×23, and every leg goes ×23 to ×31 because H = Σ2/q grows from 1.7857 to\n2.0526 and the pair sum from 0.39343 to 0.52243 on top of W. K goes 435 → 1739,\nwhich touches only the O(K²) legs, and those are milliseconds. DERIVED.\n\nMeasured @19 per leg on the timed scratch copy: `level` 0.03 s, `part0` 0.00,\n`uform` 0.14, `part2` 0.01, `part3` 0.04, `part4` 0.32, `part5j2` 0.16, total\n0.70 s. Work-scaled to @23 leg by leg: 0.69 + 3.70 + 0.98 + 8.45 + 4.90 =\n**18.7 s**. MEASURED, then DERIVED.\n\nWork-scaling is only honest if the memory hierarchy does not punish the larger\narrays, which grow from about 240 MB at @19 to 5.6 GB at @23. The probe timed\nthe two dominant kernels at both sizes: the u-form increment goes 3.9 → 5.9\nns (×1.5) and the part4 factor-extraction iteration goes 21.7 → 23.9 ns (×1.1).\nConstant-stride access stays prefetcher-friendly, so the penalty is small.\nPredicted window **19 to 28 s**, ceiling with a 3× safety factor 75 s. Peak\nmemory from the array inventory, 25 bytes per slot × W, **5.58 GB** on a 64 GB\nmachine. MEASURED.\n\n**Verdict: affordable, by a factor of about 760 against the 4-hour ceiling.**\n\n**Outturn.** @23 level time **18.878 s**, peak RSS **5.69 GB**, whole file\n**94.6 s**. The derivation was out by 1%, the memory inventory by 2%. MEASURED.\n\nThis also grades reading 10 of the file's 2026-08-18 tail, which estimated\n\"tens of seconds, not hours\" before the array was edited. CONFIRMED.\n\n## 4. Pre-registration\n\n**Honesty note first, because the campaign rule exists for this.** The headline\nnumber printed on the pricing run before this file was written, so the bands\nbelow are a stated criterion, not a blind forecast. What can be checked is that\nthe arithmetic producing them uses only the four points on record and cannot\nhave been steered by the answer; it is reproduced in full so a reader can\nverify that.\n\nFrom the four points, with lnW = 7.745, 10.310, 13.143, 16.088 and lnW(23) =\n19.223:\n\n- increments +0.2421, +0.1434, +0.0657; ratios 0.592, 0.458\n- slope per unit lnW 0.09439, 0.05062, 0.02231; ratios 0.536, 0.441\n\nContinuing the increment geometrically at the last ratio gives +0.0301 and a\nfifth point of 0.2526; continuing the *ratios'* own decay gives +0.0233 and\n0.2458. The slope route gives 0.2533 and 0.2478 the same two ways. So a\nsaturating four-point extrapolation lands in **0.246 to 0.253**.\n\n| band | fifth point | reading |\n|---|---|---|\n| BROKEN | ≤ +0.2225 | the monotone rise ends |\n| FLATTENING | +0.2225 to +0.2450 | increment below the geometric continuation |\n| LEAD CONTINUES, ON TREND | +0.2450 to +0.2550 | the extrapolation's own window |\n| LEAD CONTINUES, SATURATION NOT BITING | > +0.2550 | increment above the continuation |\n\nThe monotone lead predicts a fifth point above +0.2225. Anything at or below\nthat would be the first turn in the statistic since @11.\n\n## 5. Readings\n\n**R1 [MEASURED] The fifth point is +0.2658.** Band four: the lead continues and\nsaturation is not yet biting. The increment is +0.0433 against the predicted\n+0.023 to +0.030.\n\nFull row, from the embedded block:\n\n| x | lnW | S(0) | S̄ | S_CRT | β | S_CRT/S̄ | S(0)/S̄ | X(0)/X̄ | (X̄−X(0))/S̄ | classical |\n|---|---|---|---|---|---|---|---|---|---|---|\n| 23 | 19.223 | 597,475 | 815,732.55 | 669,028.80 | 0.8930 | 0.8202 | 0.7324 | 0.9661 | **+0.2658** | 0.8884 |\n\nwith X(0) = 6,179,192, X̄ = 6,395,995.55, X_CRT = 6,249,291.79, N̄ = 5,301,450,\nK = 1739 (29..14929), H = 2.0526, W = 223,092,870.\n\n**R2 [MEASURED] The increment ratios are not monotone, so no saturating law\nfits.** 0.592, 0.458, **0.659**. Per unit lnW the slope ratios read 0.536,\n0.441, **0.619**. The four-point sequence invited a clean geometric decay and\nthe fifth point refuses it. \"The lead continues\" and \"the lead is decaying to a\nlimit near 0.26\" are different claims, and only the first survives.\n\n**R3 [MEASURED] X(0)/X̄ rises for the second consecutive level.** 1.4541,\n0.9908, 0.9482, 0.9558, **0.9661**. P5 predicted this flattening back toward 1\nand now has two rises rather than one. The paper's §10 sentence about a\ncrossing followed by a turn has a second point of turn and can drop its\n\"one turning point on four points\" hedge.\n\n**R4 [MEASURED] The factorisation closes and the truncation factor keeps\nfalling.** S_CRT/S̄ = 1.0271, 0.9788, 0.8978, 0.8416, **0.8202**, monotone at\nfive points. β·(S_CRT/S̄) = 0.8930 × 0.8202 = 0.7324 = S(0)/S̄ exactly as\nprinted. The classical column reads 0.8884 against β = 0.8930, residual\n+0.0046.\n\n**R5 [MEASURED] The @23 β leg is a reproduction gate, and it passed.**\nS(0) = 597,475, S_CRT = 669,028.80, β = 0.8930 match `paper/anchored-note.md`\n§3's @23 row and `research/OBSERVATIONS.md`'s exact table (0.8930483 against\nclassical 0.8884421, residual 4.606e−3). So @23's survivor leg was already on\nrecord and this run recomputes it by a different path. **What is new at @23 is\nthe rotation-ensemble side**, S̄ = 815,732.55 and X̄ = 6,395,995.55, which no\nother instrument computes.\n\n**R6 [MEASURED] σ_X is not computed at @23.** Nor is any z-score, nor a custody\nline. Same gate as @19: `part0` runs only for x ∈ {11,13,17}. Reading 7 of the\nfile's tail now covers two levels. The 8-to-15-σ_X framing of the annihilation\nrequirement is still a @13-and-@17 measurement and has never been extended.\n\n**R7 [MEASURED] The m-multiplicity decomposition, and the migration.**\n\nX-gap anchored − ensemble = **−216,803.55** = m=2 cell **−40,916.72** + m≥3\ncells **−175,886.83**. Per cell, (k−1)·[anchored − ensemble]:\n\n| k | anchored | ensemble | ratio A/E | (k−1)·[A−E] |\n|---|---|---|---|---|\n| 2 | 1,369,538 | 1,410,454.72 | 0.971 | −40,916.72 |\n| 3 | 817,704 | 1,063,505.93 | 0.769 | −491,603.86 |\n| 4 | 678,479 | 575,779.43 | 1.178 | +308,098.70 |\n| 5 | 152,659 | 223,074.25 | 0.684 | −281,660.99 |\n| 6 | 90,320 | 45,135.78 | 2.001 | +225,921.08 |\n| 7 | 12,301 | 2,159.12 | 5.697 | +60,851.26 |\n| 8 | 393 | 25.32 | 15.52 | +2,573.79 |\n| 9 | 2 | 0.03 | 66.7 | +15.79 |\n\nThe m ≥ 3 share of the gap runs **10.5% @17, 23.5% @19, 81.1% @23** (at @11 and\n@13 the two channels have opposite signs and a share is meaningless). The k=3,\n4, 5 ratios are stable across two levels, 0.789/1.185/0.698 at @19 against\n0.769/1.178/0.684 at @23, while the tail excess grows without bound.\n\n**R8 [MEASURED] P4's surviving half fails at @23.** At @17 the m ≥ 3 cells were\n\"large gaps that mostly cancel\": −59.24 left of a −751.28 largest cell, 8%. At\n@23 they leave −175,804 of a −491,604 largest cell, 36%. Cancellation is\nweakening as the channel takes over.\n\n## 6. The candidate mechanism at m ≥ 3, tested\n\nTODO item X names the candidate: *anchored classes {0, −2} put multi-prime\ncoincidences at multiples of products, which are dense at the origin*. Read\nliterally that is a claim of **dependence between the two strike orientations\nat the anchor**: a product q₁q₂q₃ that sits near the origin should show up in\nboth the r class and the r+2 class together, and the anchored spectrum should\nthen carry excess at large k beyond what its own two marginals allow.\n\nThe test is the anchored analogue of part3's convolution test, which the file\nonly ever runs on the ensemble. Build the anchored joint law over the natal set,\ncnt(i, j) = #{r ∈ N : ω_s(r) = i, ω_s(r+2) = j}, take its marginals, and compare\nthe true anchored spectrum n_k(0) against the independent product of those\nmarginals.\n\n| x | X(0) exact | X from independent marginals | rel err | m=2 residual | m≥3 residual |\n|---|---|---|---|---|---|\n| 11 | 28 | 27.09 | −3.254% | +7.29 | −6.38 |\n| 13 | 452 | 453.31 | +0.290% | +26.90 | −28.21 |\n| 17 | 10,381 | 10,390.29 | +0.089% | +277.90 | −287.19 |\n| 19 | 236,625 | 236,515.60 | −0.046% | +2,461.98 | −2,352.58 |\n| 23 | 6,179,192 | 6,182,284.06 | **+0.050%** | +25,287.24 | −28,379.29 |\n\n**[REFUTED] The candidate is not there, at the resolution X can see.** At @23\nthe independent-marginal model reproduces X(0) to 0.050%, against the ensemble's\n3.39% miss: a factor of 68 better. Of the −175,887 the m ≥ 3 channel loses\nagainst the ensemble, only −28,379 survives against the anchor's own marginals,\nso **84% of the m ≥ 3 gap is the one-sided cofactor law and not a coincidence\nstructure at the origin**. The two orientations at the anchor are, if anything,\n*more* independent than the ensemble's are: on n_0 at @23 both models\nover-count, the ensemble's by +0.81% and the anchor's by +0.52%.\n\n**[MEASURED] What the closed form actually is.** Part4's one-sided law already\nhas one: #{r ∈ N : ω_s(r) = i} = p_nat·(Π_i + room_i), with ratios 0.999, 1.001,\n1.000, 0.998 at i = 1..4 at @23 and p_nat = 0.14526. Convolving that law with\nitself gives X(0), and hence (X̄−X(0))/S̄, without any m ≥ 3 coincidence input.\nThat is a composite closed form for the channel, and it is the cofactor\ntrichotomy read twice rather than new arithmetic at the origin.\n\n**[INFERRED] Where the candidate could still be hiding.** X is a\nfirst-moment-weighted statistic and it weights the bulk. In the extreme tail\nthe independent model is visibly worse than in the bulk: k = 7, 8, 9 read\n0.938, 0.750, 1.340 against 0.98 to 1.02 at k = 2..5. Two of those cells hold\n393 and 2 slots. If the origin-density mechanism exists, that is where to look,\nand X will not find it.\n\n## 7. What this does not show\n\n- **Five points is still not a law.** The increment ratios are non-monotone\n (R2), so nothing here fits a limit, and Assumption A needs the statistic below\n 1−ε at *every* level. Five values below 0.27 is a lead with one more point on\n it, and the value is still rising.\n- **No σ_X, no z-scores, no custody sweep at @23** (R6). Anyone quoting @23 as a\n fifth point of cap-31's \"pair statistics cannot see this\" argument would be\n quoting a number that was never computed. That argument remains @17-only.\n- **The @23 ensemble column is unverified against brute force.** The u-form is\n checked against the sweep at @11, @13 and @17 and rests on the identity\n holding below at @19 and @23. Its survivor leg is separately confirmed (R5);\n its X̄ and S̄ are not.\n- **P1 is not tested at @23.** `part5(·,3)` is gated at x ≤ 17, so the anchored\n aligned-versus-mixed j = 3 census, which is the direct instrument for the\n origin-density claim, has no @23 row. §6 refutes the candidate through\n independence, which is a different and weaker cut than the census would be.\n- **The refutation in §6 is about X, not about the spectrum.** Individual cells\n still miss the independent model by ±25,000 to ±29,000; what cancels is the\n total. A statistic weighting the cells differently would see structure X does\n not.\n- **No sixth point from this instrument.** The per-slot array inventory is 25\n bytes (A 1, a and b 2 each, C 4, om 2, sp 8, isP and its sieve 2, piC 4). @23\n needs 5.58 GB and measures 5.69 GB. @29 has W = 6,469,693,230 and needs\n **162 GB**; its CPU cost would be about ten minutes. The wall is memory, and\n a segmented or streaming rewrite clears it.\n\n## 8. Files touched\n\n- `research/natal-cap-35-x-multiplicity.js`: levels array `[11,13,17,19]` →\n `[11,13,17,19,23]`; tail re-embedded via `research/qc/embed.js --streams both\n --timeout 600`; a dated READINGS ADDENDUM appended below the block (it sits\n after the OUTPUT banner, so the `code-sha256` binding is untouched, and\n `--check` confirms both hashes still match).\n- `research/history/staging/xchannel-at23.md`: this file.\n\nNo live document was edited. TODO.md item X, `paper/anchored-note.md` §10 and\n`research/moire-theorems.md` all carry four-point sequences that R1 to R4\nsupersede, and none of them were touched.\n", "research/QUESTIONS.md": "# Questions: what has been asked here, what it got, and where the record is\n\n*(GENERATED from the `` blocks at the top of the notes; never\nhand-edited. Regenerate with `node research/qc.js --index` or\n`node research/gen-questions-index.js`. The block format and the gate that\nenforces it are documented at the top of `qc/questions.js`.)*\n\n**Before proposing an attack or briefing an agent:** find the TODO item in\nsection 1, search for the actual object, then read the relevant records and\ntheir decisive dependencies. Use [OUTCOMES.md](OUTCOMES.md) for reuse and\nrevisit conditions. An item can span many unrelated attempts; loading every\nlinked note is not the required first step. Verify any conclusion that\ndetermines the next move, including an ANSWERED or CLOSED entry.\n\n**Status describes the question, not proof strength or review grade.**\nOPEN: unresolved. PARTIAL: specified parts are resolved. ANSWERED: the bounded\nquestion has an answer, which may be a negative finding, finite measurement\nor completed specification. Its underlying arithmetic target may remain OPEN.\nCLOSED: the stated premise or approach is closed at its recorded scope; this\ndoes not refute every related conjecture. SUPERSEDED: another record replaces\nthis one. Consult the owning note and linked reviews for HELD or other grades.\n\n**Multiple records:** conflicting statuses are shown as MIXED. The displayed\nverdict comes from the last record in path order, not necessarily the newest\nor most authoritative review. Read the linked records before resolving a\ndisagreement; the generator does not adjudicate it.\n\n## 1. By TODO item: what has already run under each\n\n| TODO item | question | status | verdict | records |\n|---|---|---|---|---|\n| C | `Q-agent-start` What must an incoming agent read and preserve before continuing the reviewed arithmetic campaign? | ANSWERED | Onboarding and document ownership only. The second dispatch is completed and independently reviewed on 2026-09-09. RESEARCH-EXECUTION.md owns current dispositions and candidate next obligations; RESEARCH-HANDOFF.md owns the updated exact contract. No assignment is started by that review. Twin-prime infinitude and required signed estimates remain OPEN. | [AGENT-START.md](AGENT-START.md) |\n| C | `Q-bilinear-fold-first-attack` Does assigning each Mobius factor to its first fold, or applying a direct second-moment bound, establish the missing one-sided estimate? | ANSWERED | The first-fold identity yields no saving; a direct second moment has a negligible diagonal and an open two-prime off-diagonal. Fixed-fold covariance is an exact residue-imbalance square and the beta weights balance at each fixed modulus. Growing-depth joint control remains OPEN; B>=-C2*x+o(x) is not improved. | [bilinear-fold-attack.md](bilinear-fold-attack.md) |\n| C | `Q-centered-discrepancy-estimate` Can the top of D_y be removed using ordinary prime distribution after an exact divisor flip, and what estimate remains? | PARTIAL | Equations (3)-(4), the exact split and flip, survive review and finite controls. A repair of the top-range bound T^top=O_(A,eps)(x/log^A x) is DERIVED in section 3a (2026-09-08) with the endpoint atom retained, odd square divisors, the corrected reciprocal-totient local factors with (m',g)=1, explicit truncation powers, a Cauchy multiplicity device, and a uniform Mobius mean proved from the q=1 Siegel-Walfisz statements; its inputs are the terminal-point BV of Tao Notes 3 Theorem 17 with the prefix form derived by rounding. The handler (lane V) read the derivation on 2026-09-08, reconstructed (BV*), (3a.9), (3a.14) and (3a.16), checked the limiting constant of (3a.9) numerically at eight (b,g) pairs and found no defect; it is accepted at its stated scope, with ineffective constants. The payoff is D_y=D^(e_1)+O_(A,eps)(x/log^A x) only, so the handoff consumer is equivalent to D^(e_1)>=-4x/25+o(x). The fixed-endpoint signed discrepancy, D_y>=-4x/25+o(x) and twin-prime infinitude remain OPEN. | [centered-discrepancy-estimate.md](centered-discrepancy-estimate.md) |\n| C | `Q-centered-discrepancy-measurement` At reachable x, is the OPEN sufficient input D_y>=-4x/25 numerically violated, is the absolute form (13) numerically plausible, and does D_y carry structure beyond a random-sign model? | ANSWERED | MEASURED to x=2^38. Neither pre-registered falsifier fires. D_y/x stays within 0.0043 of zero for the measured j>=26. Over 30<=j<=38, D_y/x differs from -(T1/x-C2) by at most 2e-4, with maximum 1.84401e-4 at j=30; the classical term dominates this finite comparison. The data do not establish an asymptotic fluctuation scale, an impossibility of informative future measurements, or the sufficient signed estimate. No proof status changes. | [centered-discrepancy-measurement.md](centered-discrepancy-measurement.md) |\n| C | `Q-chain-review-0906` Do the written derivations from S(x) down to E_dagger hold as stated, with every imported theorem used inside its hypotheses and every discarded term at the claimed O_H(x/log^H x) rate, and where does the chain remain unreviewed? | ANSWERED | Three independent adversarial passes found no defect in the six classical contributions from S(x) to E_>, none in the grouped moment (1)-(20) including the E_* to E_dagger step, and none in the residual-coverage link E_> to E_*, whose block-shape derivation was reconstructed and verified exactly. Every imported theorem was confirmed in its primary source except the Mobius Bombieri-Vinogradov citation, which points at an exercise sheet and needs a published replacement. Stale o(x) rates in three note bodies were replaced. A checked argument is not a refereed theorem; the OPEN signed margin is untouched. | [chain-review-0906.md](chain-review-0906.md) |\n| C | `Q-chen-fold-benchmark` Can the fold framework reproduce a classical Chen lower bound on long intervals with every imported hypothesis and error budget stated? | ANSWERED | The classical implication is derived below using three explicitly imported theorem statements, with the combinatorial minorant proved and its integral margin certified by exact rational arithmetic. This is a benchmark, not a new prime theorem, and it does not prove twins or a short-zone result. | [chen-fold-benchmark.md](chen-fold-benchmark.md) |\n| C | `Q-chen-opportunity` Does the signed Chen target identify an approachable proof mechanism, and what should the next research obligation be? | ANSWERED | The target is sufficient but subtracts an avoidable odd-composite penalty; even cancellation of its obvious signed components does not close the existing separate bounds. Select a prime-detecting weight against explicitly justified arithmetic input ranges before pursuing a new estimate. The opportunity is a research judgment, not a proved mechanism. | [chen-opportunity-audit.md](chen-opportunity-audit.md) |\n| C | `Q-chen-signed-target` What exact additional arithmetic estimate would turn the completed Chen benchmark into a proof of infinitely many twin primes? | PARTIAL | An elementary prime minorant gives a conditional twin theorem with all weights, ranges, quantifiers and errors specified. The opportunity audit identifies an avoidable negative-weight penalty and a separate-estimate budget that fails even under component cancellation; this sufficient target remains OPEN and is not the sole next proof obligation. | [chen-signed-target.md](chen-signed-target.md) |\n| C | `Q-coefficient-structure` Does the arithmetic structure of the aggregated coefficients reduce the endpoint estimate beyond the previous rectangle, and can smaller coefficient norms alone supply the remaining power saving? | ANSWERED | Exact classification gives a nonsquarefree L2 norm of size at most sqrt(D)*W^(1/4) times logarithms, while the squarefree top-band energy is at least a constant times D*W*log(W). At d~x^0.277, e~x^0.467 every endpoint sector except the squarefree-squarefree prime sum has a power saving. This reduces that rectangle to one explicit endpoint correlation; prime-dispersion.md supplies the remaining estimate and controls the full rectangle. Finite algebra, CRT sectors and archived prefixes are checked separately. | [coefficient-structure.md](coefficient-structure.md) |\n| C | `Q-cofactor-progression-transfer` Can the corrected full-cofactor kernel be estimated on a growing cofactor range by representing its actual arithmetic on CRT progressions, and does the resulting bound reach the twin consumer? | PARTIAL | Derived from named imports: for some positive absolute kappa and d, all prime-r cofactor pairs s,t <= floor((log x)^kappa), including non-squarefree inputs and mixed transition/smoothed profiles, admit a dyadic scale-average O(log^(2-d) X) bound after division by x. The extension uses bounded multiplicative Euler modifications on the progression s\\|n, t\\|n-2 and keeps its density 1/lcm(s,t), so summing cofactors costs only O((log log X)^2). This goes beyond s=t=1 but remains short of o(x). The large-cofactor joint remainder, outside residual and twin margin remain OPEN; the attempted extension to power-sized cofactors fails on source modulus range and accumulated logarithmic cost. | [cofactor-progression-transfer.md](cofactor-progression-transfer.md) |\n| C | `Q-consumer-comparison` How does the sufficient margin C_2 x + E_dagger(x) >= c_0 x/(log x)^K compare logically with the published sufficient routes to twin primes, and what does the literature already own of the corner object? | ANSWERED | The sufficient margin remains OPEN. The published conditional routes imply it where their hypotheses and stated lower bounds apply; reverse implications are not established. The dictionary is E_dagger(x)/x = C_2(2 delta_x - delta_{x/2} - 1) + o(1), with K>0 below that normalisation's resolution. Independent review corrects the comparison: cumulative and dyadic consumers on unbounded scales are equivalent at the same fixed K with possibly different positive constants. At K=0 an eventual cumulative lower bound also implies the unbounded-scale dyadic consumer. The nonnegative-weight corner identification applies only to the prime-cofactor subfamily; the full residual retains additional branches. Every surveyed route remains conditional. | [consumer-comparison.md](consumer-comparison.md) |\n| C | `Q-corner-branch-diagnostic` Can future experiments enumerate the full coefficients C(n) and C'(n-2) branch by branch, without silently dropping proper prime powers, cofactors s>1, non-squarefree inputs or either gcd branch? | ANSWERED | The driver exists and its two independent constructions of each side agree exactly on 8192 retained sides and 1802 fixture sides; five deletion and reconstruction controls are ACTIVE in both runs. Its split by gcd(n,n-2) measures integer parity, not the per-term CRT gcd partition. The fixed-eta corner is empty on the retained prefix, and Elo exceeds E0 across the entire J_x, so the matrix is a proxy window. This is finite algebra and tool preparation, with no asymptotic estimate or twin margin. | [corner-branch-diagnostic.md](corner-branch-diagnostic.md) |\n| C | `Q-corner-coefficient-energy` Can the full corner coefficients be studied before splitting their cofactor branches, and what do exact energy and classical smoothed sieve weights supply? | ANSWERED | The sharp coefficient energy is an explicit signed quadratic form in short cofactors plus O_epsilon(x^(1/2)K^2*x^epsilon), a power below x at the fixed corner. Logarithmically averaging the divisor cutoff gives Barban–Vehov weights and an O_eta(x log x) squared norm on each side and absolute shifted-product bound. The follow-up sharp-corner-transition.md gives sharp O_eta(x log^2 x) norms and proves a non-negligible norm transition for sufficiently small fixed eta. Signed transition saving and the global complement remain OPEN; no exact residual cut or twin margin changes. | [corner-coefficient-energy.md](corner-coefficient-energy.md) |\n| C | `Q-corner-correlation` On the corner S_0 where both cofactors sit just above their own cutoffs, what exactly is the remaining endpoint sum, what does the sufficient consumer need there, and does any reviewed correlation theorem supply it? | PARTIAL | The full corner is exactly sum_n C(n)C'(n-2). Proper-prime-power terms above the cutoffs are negligible at fixed eta by §1.1; the remaining prime-r terms with s>1 or s'>1 and non-squarefree inputs remain OPEN. The s=s'=1 piece has a nonnegative Mobius-product weight with exact moving cuts. The multiplicative lift and round-review sampling give small continuous and dyadic scale-average log savings for that subfamily, with no o(x), full-corner control or twin margin. The complement is still open. | [corner-correlation.md](corner-correlation.md) |\n| C | `Q-corner-log-average` What do logarithmically averaged correlation theorems and their quantitative successors actually supply for the prime-cofactor corner weight, with exact support, rates and scale quantifiers? | PARTIAL | The pure band is invariant under dilation by primes outside it; the exact cofactor window is not, but its support extends to x^(1-2eta) and need not leave the window under a fixed dilation. Uniform-prefix logarithmic bounds imply block bounds by Abel summation; a relative exponent above 3 is one sufficient route to o(x), not a universal necessary threshold. Pilatte's 1/(96e) calculation applies only to his displayed parameter choice. A bounded multiplicative Fourier lift now gives a small continuous scale-average log saving for the prime-cofactor subfamily via Tao-Teravainen v2; see prime-band-transfer.md. No o(x), dyadic pointwise estimate, full-corner estimate or positive twin margin follows. | [corner-log-average.md](corner-log-average.md) |\n| C | `Q-corner-measurement` At finite dyadic x, how large is the corner two-point correlation K(x)=sum_{x/2(x)>=c*x/log^K x on unbounded dyadic scales suffices, as does a stated logarithmically rescaled average. These implications are derived from named inputs; their endpoint hypotheses remain OPEN. The recorded uniform-gap theorem comparison has different quantifiers and does not establish an obstruction for this consumer. | [endpoint-target-audit.md](endpoint-target-audit.md) |\n| C | `Q-fixed-endpoint-discrepancy` After the accepted truncation to odd moduli e=-4x/25+o(x)? | PARTIAL | Reviewed 2026-09-09 after the coprimality repair: T_I^low=O_(A,eps')(x/log^A x), hence S=C_2x+B+O_A(x/log^A x), with B the exact Type II plus band sum (2.9). The untruncated modulus bound was false; g<=(log x)^(A+13), both tails, the coprime density and the weighted BV multiplicity now pay the estimate. Review corrects harmless log powers and the power-of-two atom over the full eps' range. D^(e_1)>=-4x/25+o(x) is equivalent to B+2C_2M>=-4x/25+o(x); it implies, but is not equivalent to, B>=-(C_2-1/200)x+o(x). The latter and H_B are sufficient OPEN margins. The absolute band statement (4.9) is one stronger sufficient input, not a necessary condition. No inspected source estimates the remaining actual signed B. Twin-prime infinitude remains OPEN. | [fixed-endpoint-discrepancy.md](fixed-endpoint-discrepancy.md) |\n| C | `Q-fold-arithmetic-bridge` Does one parity-table bridge from the anchored fold ledger yield a sufficient twin lower bound with named arithmetic inputs? | PARTIAL | Exact identities retained. The one unread sieve input of the pricing, Bombieri--Vinogradov for k-fold X^(1/u)-rough products, is derived from Wu's Lemma 2.3 (section 3a); the same input, sieved in the composite variable, gives the contamination aggregate constant 4 for each fixed k,u, replacing the displayed 12.86 to 19.72. With these inputs, elementary bounds give Q_cov(u)<1 and c*_real(u)<4 for every u>4 (section 4a, independently reviewed 2026-09-09 with rational certificates), so neither sufficient ratio test succeeds at any depth. This closes the two tests, not the decorrelation hypotheses, and supplies no twin estimate. | [fold-arithmetic-bridge.md](fold-arithmetic-bridge.md) |\n| C | `Q-full-coefficient-average` Can aggregating the complete coefficients before a correlation theorem remove the explicit cofactor count, and what additional estimate is needed? | PARTIAL | Exact factor and rounded-endpoint Fourier identities retained, with c_(i,0)=3/5 and an explicit composite-filtered weighted sum. The full family is not 1-bounded, but the sufficient phase condition admits at least k=0,+/-1 on the left and l=0,+/-1,...,+/-6 on the right for every Mellin twist; the earlier zero-only claim is corrected. Composite filtering and the required correlation rate remain unmatched. Proposition 6.5, independently reviewed including on 2026-09-09, proves coefficient norm at least (log x)^(2/5) for representations by 1-bounded functions on all smooth inputs. This does not exclude density-one representations, paid growing components or a jointly treated Fourier sum. No sufficient signed twin margin follows. | [full-coefficient-average.md](full-coefficient-average.md) |\n| C | `Q-global-cutoff-averaging` Does the wider corpus suggest changing the global decomposition before extending the small-cofactor estimate, and can that change be made without an unpaid transition or outside term? | PARTIAL | Derived: independently averaging the two initial Mobius cutoffs over fixed exponent intervals gives S=C2*x+sum G_L(n)G_R(n-2)+O_A(x/log^A x), uniformly on dyadic scales, with all cofactor and prime-power branches included. Each full G has squared norm O(x log x). The existing sharp-corner lower norm then forces cancellation between the corner coefficient and the rest at the same input. This does not estimate their shifted product. The global signed residual was already O(x) by the sieve upper bound and positivity; the new norm representation is not an improved signed bound. Prefer a bounded attempt on this global coefficient pair and a one-sided consumer; the twin margin remains OPEN. | [global-cutoff-averaging.md](global-cutoff-averaging.md) |\n| C | `Q-global-factor-signs` Which full factor configurations cause negative global products, can prime-power exceptions be paid, and does a pair-trigger upper bound control the negative part? | PARTIAL | Derived: G_i(n)=Lambda_(>W_i)(n)-F_i(s_i(n))*log t_i(n)-E_i(n), with all small-prime factors in s_i and all primes of t_i exceeding W_i. The correction is supported on a small-prime power exceeding W_i; its full shifted effect is O_epsilon(x^(39/40+epsilon)). After also paying proper prime powers, the residual is a composite-filtered signed smooth/rough cofactor sum. A ten-small-prime cell has F=-84 although every pair lies below a, refuting the proposed pair-trigger majorant. Prime filters contribute explicit terms when bounding the negative part; dropping them is an upper bound, not an identity. No scale-x signed improvement or twin margin follows. | [global-factor-signs.md](global-factor-signs.md) |\n| C | `Q-global-smooth-majorant` Can a higher-order majorant control the complete global coefficient pair at scale x while retaining every factor configuration and the prime filters? | PARTIAL | Derived using the corrected Henriot upper theorem: a C3 probability average of the same admissible initial cutoffs gives a full residual with sum \\|Ghat_L(n) Ghat_R(n-2)\\|=O(x). The three-smallest-prime majorant has bounded harmonic mass and meets the source growth class uniformly, despite not being multiplicative. All prime-power exceptions are paid. This replaces the O(x log x) absolute budget for the earlier logarithmic profile by O(x) for a different admissible profile representing the same signed residual to arbitrary logarithmic precision. The implied constant is not compared with C2 and no improved signed lower bound or twin margin is supplied. | [global-smooth-majorant.md](global-smooth-majorant.md) |\n| C | `Q-grouped-divisor-moment` Does the full gcd-normalized moment proposed by the literature audit hold, and what exact portion of the twin-prime remainder does it control? | ANSWERED | The proposed moment is derived from classical completion with all coefficient sectors and uniform twists included. Full rectangles are controlled when delta<19/25 and delta+3nu<161/100, in addition to the preceding region. A concrete extra cut d<=floor(x^(151/200)), de^3<=floor(x^(321/200)) controls the entire d~e~x^(2/5) benchmark and leaves an explicit smaller-domain endpoint remainder. The uniform product threshold stays below 19/25. At delta=8/25,nu=9/20, the next deficit is confined to small-common-divisor nonzero kernels; their required saving and the global twin margin remain OPEN. Finite validation does not prove asymptotic rates. | [grouped-divisor-moment.md](grouped-divisor-moment.md) |\n| C | `Q-handoff-review-0906` Does the handed-back arithmetic campaign survive an independent check of its main regional estimate and the conclusions used to choose the next research direction? | ANSWERED | The bounded handoff audit is completed in reports 20 and 21: the checked local reduction and regional mechanisms survive, named source statements were verified, and the consumer, corner support, rate, shrinking-margin and identity-piece overclaims were corrected. Joint Cauchy is now priced and adds no region. The multiplicative band transfer has a separate PARTIAL owner with a weaker continuous-scale payoff. Imported deep theorems remain imports; this is not corpus-wide certification or a twin margin. | [handoff-review-0906.md](handoff-review-0906.md) |\n| C | `Q-heath-brown-edges` Does the a=1 edge, and hence the corner S_0, come from the particular two-cutoff Vaughan identity in prime-detection-spec.md and shifted-prime-decomposition.md, or does an equivalent edge reappear under Heath-Brown's identity and the other standard decompositions of Lambda? | ANSWERED | The exact identities and exponent simplex leave a=1 in separately bounded formal pieces, including Heath-Brown's balanced j=K configuration. This does not prove an intrinsic edge after identity-piece cancellation. The bounded-cofactor object has affine dilations, not literally F(n)F(n-2). Linnik's fixed truncation has a large term-wise absolute tail by an elementary counting proof; its signed tail is not thereby bounded below. Heath-Brown K=3 narrows the classical split band, with coefficient and Type I normalization obligations retained. No full alternative reduction, new region or twin margin follows. | [heath-brown-edges.md](heath-brown-edges.md) |\n| C | `Q-joint-correction-source-audit` Does completing the supported prime-partner correction yield cancellation beyond the existing Vaughan reduction, and which literature inputs actually justify that completion? | ANSWERED | CHECKED prior-art matches for subset convolution, complementary sieve sums and signed smooth-number distribution. DERIVED reconstruction identifies the unrestricted factor weight with an existing short divisor approximant, up to the paid prime-power error. Its supports are x^(12/25) and x^(1/10). Reassembling the entire joint correction returns the already-open Vaughan remainder and supplies no new signed bound. A formal factor-vector control distinguishes the unrestricted logarithmic weight from the selected prime-cofactor branch. The matched general sieve theorems require inputs not supplied here. Literature novelty is unestablished; twin infinitude remains OPEN. | [joint-correction-source-audit.md](joint-correction-source-audit.md) |\n| C | `Q-joint-factor-estimate` Can an additional estimate for actual factor configurations control the full signed balance of the C3 residual? | ANSWERED | DERIVED: for the single-large-prime delta-rough family F, the residual splits exactly as R_F = C2*x*H_delta - Z_F^p + o(x), where the Type I part C2*x*H_delta is evaluated by ordinary prime BV on each cofactor interval and the twisted Mobius constant 2C2 (corrected 2026-09-08 after V2: Wu Lemma 2.3 is squarefree-only and did not cover the non-squarefree b_R moduli; prime BV covers all moduli), and Z_F^p is the signed prime-partner sum. The complement carries the opposite Type I part -C2*x*H_delta, so grouping by factor configuration transfers Type I mass and leaves the twisted prime correlation B_L unchanged. With the matched upper sieve the family bound is C2*x(1+H_delta-(40/9)H^+), certified below -2/5 C2*x: INSUFFICIENT BOUND. Even a sieve constant 2 in place of 40/9 needs H^abs<1 and leaves the complement unpaid. Independently reviewed 2026-09-09 with the uniform delta-range and complement corrections below. Twin infinitude and every sufficient margin remain OPEN. | [joint-factor-estimate.md](joint-factor-estimate.md) |\n| C | `Q-kernel-sign-control` At delta=8/25, nu=9/20, is the moment Mfrak and its small-common-divisor kernel X_small measurably smaller with the actual coefficient b_u=A_right(gu) than with \\|b_u\\| or with \\|b_u\\| times an independent seeded random sign, and does the actual/random ratio decrease with x over the feasible dyadic range? | ANSWERED | MEASURED, negative. Over seven dyadic scales j=18..30 and four boxes, the pre-registered slope of log2(actual/random) for X_small has mean + sd >= 0 in all four families, so the run gives no heuristic support for extra Mobius-sign cancellation in the small-j kernel. The actual coefficient's \\|X_small\\| ranks like one more random draw. The measurement is weak where it matters: the finite moment is 97.66-102.58 percent equal-frequency class, X_small carries 0.02-2.88 percent, J0=floor(x^(1/20)) is 1 at j=18 and 2 at j=20..30, and Z=2 collapses the prime-power sector of (13) to r=2. Equation (21) and the global twin margin remain OPEN and untouched. | [kernel-sign-control.md](kernel-sign-control.md) |\n| C | `Q-left-divisor-signs` Can the Möbius structure of the left coefficient A_left(gm) replace the first Cauchy inequality, so that its sqrt(M) loss is not paid, and what does that buy at the target box (delta,nu)=(8/25,9/20) and at the corner a=b=1? | PARTIAL | Partly, and not where it is needed. Two lemmas are derived: a Type I bound with no zero-frequency budget, usable at the target box whenever the arbitrary block is shorter than x^(1/4), and a Type II bound that is uniform in the Perron twist heights. Applied to the actual coefficient they confine the target-box deficit to sectors with left prime power r>=x^(43/200) and right prime power q>x^(3/100), and lower the worst block exponent there from 103/100 to 41/40. The corner a=b=1 is untouched: grouped, Type I, Type II, Bettin-Chandee/Wright and even a hypothetical per-block square-root bound all exceed 1 there. No region is added, E_dagger and the consumer (20) are unchanged, and the global twin margin remains OPEN. | [left-divisor-signs.md](left-divisor-signs.md) |\n| C | `Q-mobius-bv-derivation` Is there a published theorem statement of Bombieri-Vinogradov for the Mobius function at level T^(1/2)/log^B T with a maximum over coprime classes, and if not, does that estimate follow from published theorems by a derivation whose every hypothesis is checked? | ANSWERED | No published theorem statement was located in five channels searched in the owning convention named in research/SEARCH-CONVENTIONS.md; the closest is Granville-Shao's published assertion that the result is known, which carries no proof locator for the Mobius case. The estimate is derived here from four numbered results of Koukoulopoulos, GSM 203 (Corollary 13.4, Theorem 26.2, Theorem 26.6, equation (26.3)) plus Vaughan's identity for mu, with B(A)=A+6 and an ineffective constant. This is a derivation from published theorems, not a refereed theorem, and it is read from the author's preliminary version of that book. | [mobius-bv-derivation.md](mobius-bv-derivation.md) |\n| C | `Q-moving-cutoff-parity` Does a proof-level reading of the shifted-Mobius sieve give a valid, explicit error-tolerant arithmetic target, and does its divisor switching preserve the moving boundary? | ANSWERED | CHECKED prior art: Murty--Vatwani already supplies the conditional squarefree/parity mechanism and fixed-residue reduction. Its printed p. 654 swap omits n+h>ey; an exact finite counterexample verifies failure of that equality. DERIVED dyadic repair keeps the boundary and gives S=C2*x-2*C2*M+D_y+O_A(x/log^A x). A one-sided D_y>=-4*x/25+o(x) on unbounded dyadic scales is sufficient, using the certified margin C2*(1-A2)>33/200. That discrepancy estimate remains OPEN. Vatwani's recovered author preprint clarifies the one-odd-exponent cases and unfilled near-level-one hypotheses. No new signed bound is obtained; literature novelty is unestablished. | [moving-cutoff-parity.md](moving-cutoff-parity.md) |\n| C | `Q-next-correlation-source-map` Is there a primary-source statement with stronger scale or coefficient quantifiers than the ones the prime-band transfer actually consumed? | ANSWERED | Scoped negative for the inspected statements: function-class and rate mismatches remain. The coordinating review withdraws the all-scales/dyadic-omitting requirement as a necessary gate, since bounded-interval stability yields a weak dyadic scale average from the existing input. No sufficient rate or full-coefficient theorem is matched. Proper-prime-power terms are already negligible on the fixed corner; other prime-r cofactor branches remain open. | [next-correlation-source-map.md](next-correlation-source-map.md) |\n| C | `Q-next-transfer-review` Does prime-band-transfer.md (1)-(6) follow at exactly its stated scope, and does any step fail against an adversarial reading of its hypotheses? | ANSWERED | The derivation (1)-(6) survives at its stated continuous-average scope. The coordinating round review corrects check l: interval stability additionally gives a dyadic scale average, with a further exponent loss. There is no every-dyadic estimate, o(N), full-corner control or twin margin. Proper-prime-power branches have a separate negligible fixed-corner bound. | [next-transfer-review.md](next-transfer-review.md) |\n| C | `Q-paired-factor-budget` Does averaging the proposed two-small-prime loss and three-small-prime gain using existing distribution theorems make their net contribution sufficient for the current twin consumer? | ANSWERED | DERIVED from the matched classical inputs, with finite rational enclosures VERIFIED: the explicit positive family has liminf P3/(C2*x)>1, but liminf (N2-P3)/(C2*x)>11/10. Keeping only these families and the main term therefore fails at the current cutoffs. Pan--Ding supplies aggregate level x^(9/20), stronger than the earlier per-cofactor application; the complete signed margin remains OPEN. The aggregation machinery is prior art and literature novelty of this application is unestablished. | [paired-factor-budget.md](paired-factor-budget.md) |\n| C | `Q-polylog-fold-transfer` Can the fixed-fold calculation be made uniform at growing depth, and transferred to the actual bilinear remainder with an adequate error? | ANSWERED | The local bilinear model is O_H(x/log^H x) uniformly for polylogarithmic squarefree moduli. Its absolute reconstruction error is at least (1/2+o(1))*x. The needed signed reconstruction bound remains OPEN, so this model does not establish the twin-prime lower bound. | [polylog-fold-transfer.md](polylog-fold-transfer.md) |\n| C | `Q-prime-band-completion` Can the remaining squarefree correlation be localized to simpler prime and harmonic bands, and does completion then permit a direct general bilinear Kloosterman estimate? | ANSWERED | The pilot reduces, with power-saving endpoint errors, to p in (x^0.235,x^0.24], q in (x^0.045,x^0.05] and positive h in (x^0.029,T], with T=O(x^0.0342). Both primes exceed 2h eventually. Completion is exact but produces coupled h-dependent coefficients on every dual residue modulo q. Its full-frequency Kloosterman matrix has norm exactly q, so a uniform power saving by an arbitrary-coefficient operator bound is impossible at that scope. This completion alone leaves the 61/20000 budget deficit; prime-dispersion.md controls the pilot using a different second moment. No twin lower bound is established. | [prime-band-completion.md](prime-band-completion.md) |\n| C | `Q-prime-band-transfer` Can the nonmultiplicative prime-band weight be represented by bounded multiplicative functions, and what does an available quantitative correlation theorem actually give after all scale and normalization costs? | PARTIAL | A smooth Fourier superposition represents mu(n)L_B(n)/log X exactly on n<=2X+2. The quantitative non-pretentious input and a mesh give a continuous scale-average saving for the exact s=s'=1 windows. The round review additionally derives a dyadic scale-average saving by bounded-interval stability, with exponent c_*/2. Neither estimate gives an every-dyadic bound, o(N), full-corner control or a twin margin. The extension in cofactor-progression-transfer.md now estimates a growing polylogarithmic cofactor family; the full remainder remains open. Proper-prime-power branches have the separate bound in corner-correlation §1.1. | [prime-band-transfer.md](prime-band-transfer.md) |\n| C | `Q-prime-detection-inputs` Can we give a locally correct positive comparison sequence, justify the ordinary arithmetic inputs, and prove exactly what one additional estimate would buy? | ANSWERED | Yes: the comparison sequence and Type I estimates are derived, and a classical Vaughan decomposition gives S(x)=C2*x+B(x)+O_H(x/log^H x) for every fixed H. A fixed one-sided saving over B>=-C2*x+o(x) implies twins on unbounded scales. This is a fully specified classical reduction, not a new cancellation estimate or proof mechanism. | [prime-detection-spec.md](prime-detection-spec.md) |\n| C | `Q-prime-dispersion` Can a second moment retaining two right primes control the remaining localized squarefree pilot after its actual conductor and coefficient costs are included? | ANSWERED | Yes at this rectangle: aggregation in m=dp before Cauchy gives a logarithmically bounded coefficient on a long interval. The only zero-phase pairs are identical (q,h). Distinct primes give modulus e*q1*q2 with gcd loss at most O(A*Z); completion and Weil bound all off-diagonal terms. The diagonal has exponent 1989/2000 and the largest off-diagonal exponent is 1173/1250 before an arbitrarily small loss. The remaining high-frequency polynomial is O(x^(199/200)); consuming the previous reductions gives R_IJ=O_H(x/log^H x) at D=x^0.277, E=x^0.467. This is a full rectangle at product scale x^0.744, not a uniform product cutoff or a twin lower bound. | [prime-dispersion.md](prime-dispersion.md) |\n| C | `Q-prime-power-dispersion` Does the repeated-prime structure control the BB contribution beyond its two-norm bound, and can a prime-power moment include every harmonic uniformly? | ANSWERED | The decomposition leaves an exceptional coefficient with norm at most sqrt(D) times logarithms; its odd prime-power core has a derived all-harmonic dispersion bound. The same bound for first powers removes the former left-prime cutoff and unpaired diagonal constraint. Two strict inequalities control a region including d~x^0.247, e~x^0.612 at product scale x^0.859. Their product-exponent supremum is 5757/6700, not a uniform product cutoff or a twin lower bound. Exact finite checks run separately from this classical-input derivation. | [prime-power-dispersion.md](prime-power-dispersion.md) |\n| C | `Q-reachability-coverage` Which part of the original divisor domain can no improvement to the grouped moment's nonzero kernel reach, in either orientation, and what would a uniform saving on that kernel actually buy? | ANSWERED | For a prescribed fixed exponent margin eta0, the zero budgets leave S_0={p<2eta0,q<2eta0} below that required margin regardless of the nonzero-kernel saving. Uniform gamma=2 covers its complement at that prescribed margin. Smaller positive margins can reach fixed interior points of S_0; only p=q=0 remains at zero budget one for every margin. The proposed eta0~loglog x/log x substitution is not justified by estimates carrying x^epsilon losses; its box count is geometry only. The sufficient twin margin remains OPEN. Independent review F3 and F5 correct the quantifiers and the prime-power tail proof; round-review-0906 corrects the diagonal scope, fixed-margin bulk claim and six-cut separation. | [reachability-coverage.md](reachability-coverage.md) |\n| C | `Q-research-execution` How should agents use the integrated returns, avoid repeated work and select the next bounded research obligation? | ANSWERED | The second dispatch from 1285d47 closed at 46fa948 and was independently reviewed on 2026-09-09. A/A2/B/C/D/E and V/W returns are present and integrated at their owning scopes. A2's corrected Type I estimate is accepted; D's fourth residual cut is retained and its remaining moment saving corrected to 7/200. C's low nonzero modes and E's aggregate scope are corrected. No new assignment is started by this integration; next obligations are candidates, not running work. Twin-prime infinitude and every sufficient signed margin remain OPEN. | [RESEARCH-EXECUTION.md](RESEARCH-EXECUTION.md) |\n| C | `Q-research-handoff` What must an incoming research agent understand, verify and return before its work can advance the current twin-prime argument? | ANSWERED | Current handoff after the 2026-09-09 independent review of the dispatch ending at 46fa948. The full residual has four simultaneous complement conditions after D's regional estimate. A's centered truncation and A2's repaired low Type I estimate give the alternative S=C2*x+B+O_A(x/log^A x). B and all sufficient signed margins remain OPEN. The C3 absolute bound, family identities and scoped C/E failures retain their owning hypotheses. Execution is owned by RESEARCH-EXECUTION.md. | [RESEARCH-HANDOFF.md](RESEARCH-HANDOFF.md) |\n| C | `Q-research-round-validation` Which submitted results from lanes A, A2, B, C, D, E, V and W survive direct review, and what can be integrated or used to choose the next attempt? | ANSWERED | Reviewed incoming commits 07ab47f through 46fa948 on 2026-09-09. A's truncation and A2's repaired low Type I estimate are accepted with bookkeeping corrections. B's ordinary-BV family identity survives; its uniform Dickman constant and complement condition are repaired. C's norm obstruction survives on all smooth inputs, but low nonzero Fourier modes are also 1-bounded. D's block estimate and fourth residual cut survive; its next moment saving is 7/200, not 17/200. E's rough-product distribution and aggregate constant 4 are accepted at the stated Wu uniformity reading; rational certificates close the two displayed tests for every fixed u>4. W remains OCR-only at the original 1974 theorem page. All sufficient twin-prime margins remain OPEN. | [research-round-validation.md](research-round-validation.md) |\n| C | `Q-residual-coverage` What portion of the full residual is controlled with uniform margins, and can the exceptional first branch improve that coverage rather than just a regional corner? | ANSWERED | The first-branch and prime-free moments remove the exceptional norm condition. Full rectangles are controlled when delta+nu<19/25 or 5delta+2nu<123/50. Uniform twists justify exact coupled cuts: every fixed de<=x^theta with theta<19/25, and d^5*e^2<=x^lambda with lambda<123/50, as well as their union, have arbitrary fixed logarithmic savings. A concrete remaining endpoint sum has de>floor(x^(3/4)) and d^5*e^2>floor(x^(49/20)). Its twin margin is OPEN. The right zero-frequency budget, after an extra Cauchy inequality, limits the uniform product range; the left cross moment limits the other edge. Finite checks validate identities and masks, not asymptotic rates. | [residual-coverage.md](residual-coverage.md) |\n| C | `Q-review-request-0906` Which claims of the arithmetic campaign should an independent reviewer check, in what order, against which records, and what must the review return? | ANSWERED | Completed review specification at the bounded scope recorded in handoff-review-0906.md and reports 20 and 21, with the subsequent round reviewed in round-review-0906.md. The dependency list records questions at dispatch, not current unfinished assignments. Imported proofs retain their review limits and the twin margin remains OPEN. Current assignments are in RESEARCH-EXECUTION.md. | [REVIEW-REQUEST.md](REVIEW-REQUEST.md) |\n| C | `Q-round-review-0906` Which conclusions of the completed agent round survive review, and which corrections change the next research specification? | ANSWERED | The fixed-band lift and continuous moving-window estimate survive the checked source interface. Interval stability additionally gives a dyadic scale-average logarithmic saving, so the blanket scale exclusion was too strong. The rate and missing prime-cofactor terms with s>1 or s'>1 still prevent a twin margin. Proper-prime-power branches already have a negligible bound on the fixed corner. The driver computes integer parity, not the CRT gcd partition; its total coefficient identities survive. Diagonal lower-bound, reachability and six-cut bookkeeping overclaims are corrected. No region, exact residual cut or twin lower bound changes. | [round-review-0906.md](round-review-0906.md) |\n| C | `Q-sharp-corner-transition` What is the energy of the complete sharp corner coefficient, and can the smoothed estimate be transferred through a negligible L2 transition? | ANSWERED | DERIVED using named analytic inputs: for each fixed 00 both sharp squared norms and sharp-minus-smoothed squared norms are Theta_eta(x log^2 x) on the actual dyadic intervals. Thus an O_eta(x log x) sharp squared norm or negligible L2 transition fails in that range. Signed transition correlations and the global complement remain OPEN. No region, exact residual cut or twin margin changes. | [sharp-corner-transition.md](sharp-corner-transition.md) |\n| C | `Q-shifted-prime-decomposition` Does decomposing Lambda(dk-2) give a provable saving on any part of the actual shifted-prime sum, and what exact arithmetic remains outside the imported hypotheses? | ANSWERED | Both second Type I terms are O_H(x/log^H x) by classical Mobius BV. The residual is an explicit weighted sum over dk-ev=2, equivalently an average of two-linear-form Mobius correlations with growing coefficients and possibly one-point intervals. Its required one-sided improvement is OPEN. No twin lower bound or novelty is claimed. | [shifted-prime-decomposition.md](shifted-prime-decomposition.md) |\n| C | `Q-shifted-prime-mobius-sums` At reachable x, do the shifted-prime Mobius and Liouville sums, the cleanest instance of the parity input the twin reduction lacks, show any size or sign structure that a random-sign model does not, and how loose is the trivial bound on M(x) used for the centered tolerance? | ANSWERED | MEASURED to x=2^38 (10.9e9 primes). All four sums sit at random-sign size: \\|U\\|/control rms at most 2.8 at any j, slopes 0.28 to 0.65 against the control's 0.43 to 0.47 with per-draw spread 0.19 to 0.30; the pre-registered structure falsifier did not fire. M(x)/x is at most 8.8e-5 in magnitude for j>=29 against the trivial bound 0.374, so the -4x/25 tolerance in moving-cutoff-parity is loose by four orders at these scales; a proof still needs a bound on M, which is the same parity object. The squarefree density of p+2 among primes is 0.747911 at 2^38 against 2*Artin = 0.7479116. pi(2^j) matches OEIS A007053 at every j. Nothing here bears on the asymptotic conjecture. | [shifted-prime-mobius-sums.md](shifted-prime-mobius-sums.md) |\n| C | `Q-signed-divisor-grouping` Can grouping different fibers while retaining their Mobius signs control an explicit part of R, and what signed error remains after that grouping? | ANSWERED | For every fixed H>0 the actual remainder restricted to de<=x^(7/10) is O_H(x/log^H x), although each sign's mass there is at least c*x*log(x) eventually. This follows from a written classical-input derivation, with finite algebra and archived-factor checks. The remaining de>x^(7/10) contribution reduces to an explicit weighted CRT endpoint discrepancy whose required one-sided improvement is OPEN. | [signed-divisor-grouping.md](signed-divisor-grouping.md) |\n| C | `Q-signed-moment` Is there an argument for the block that does not pay sqrt(M) at the first Cauchy inequality or that extracts cancellation from the R=0 class, and what are its budgets at the corner a=b=1 and at the target box (delta,nu)=(8/25,9/20)? | PARTIAL | The true diagonal has an upper bound x^(1+o(1)) on the transition band, not a uniform nonzero lower bound. Lemma A controls only u1=u2,h1=h2; unequal proportional R=0 pairs remain outside it. Lemma B and the now-priced joint Cauchy arrangement add no region. The former general Holder floor is valid only under additional coefficient assumptions, not for arbitrary sparse coefficients. Conditional off-diagonal budgets and the random-matrix ceiling remain conditional and heuristic. No necessity of using both signs, full-corner estimate or sufficient twin margin is established. | [signed-moment.md](signed-moment.md) |\n| C | `Q-singleton-fiber-audit` Can the one-point fibers of the coupled remainder be discarded or controlled in absolute value at scale x, and what do the actual weights and archived intervals show? | ANSWERED | The grouped coefficient is zero whenever either integer is prime. Each sign's ungrouped singleton mass is at least c*x*log(x) eventually for an absolute c>0, by a written derivation from named classical inputs. Finite full-interval and archived-prefix checks validate the partition. The signed singleton and longer-fiber contributions remain OPEN; no twin lower bound or novelty is claimed. | [singleton-fiber-audit.md](singleton-fiber-audit.md) |\n| C | `Q-small-divisor-kernel` Does reciprocity separate the j<=x^(1/20) kernel of grouped-divisor-moment (21) at delta=8/25, nu=9/20, and do the Duke-Friedlander-Iwaniec / Bettin-Chandee and Kuznetsov interfaces then supply the required saving? | ANSWERED | The separation is exact and cheap: reciprocity splits index by index, the correction phase has sup norm and total variation O(A/(MN))=O(1/x) at this box, and Mellin separation of the endpoint weight costs x^epsilon with L1 mass min(1,Ax/(MN)), so the whole block reduces to a pure trilinear Kloosterman fraction with unimodular twists. The interfaces then fail by explicit amounts: Bettin-Chandee Theorem 1 gives block exponent 129/125, above the required 1 by 4/125 and above the classical 103/100; DFI (1.1) gives 1267/1200; every box Bettin-Chandee controls is already controlled, verified on 19005 rational boxes; the audit's spectral diagnostic factor is trivial for all common divisors up to x^(19/50). Closing this box inside the same interface needs the (M+N) exponent below 27/140 or the (AMN) exponent below 9/28. Target (21) and the global twin margin remain OPEN; nothing is added to the controlled region. | [small-divisor-kernel.md](small-divisor-kernel.md) |\n| C | `Q-smooth-sieve-literature` Which nearby published results should be imported instead of rediscovered, and do their methods give a more focused next question for the complete global remainder? | PARTIAL | Existing sources supply the finite-difference mechanism, one-point concentration and quadratic profile optimization; shifted divisor-sum estimates admit fixed shift two under their own support and smoothness conditions. A derived application of corrected Henriot bounds gives small-prime mass O_sigma(delta^sigma x)+O_epsilon(x^(39/40+epsilon)), uniformly in delta for fixed 0floor(x^(77/100)) as a fourth simultaneous condition of W_dagger through the Perron machinery, changing only the summation domain of E_dagger. The target box (407/400), the corner, the uniform product threshold 19/25, the sign of E_dagger and the global twin margin are unchanged; the margin remains OPEN. | [structured-dispersion-estimate.md](structured-dispersion-estimate.md) |\n| C | `Q-supported-coefficient-dickman` Can the whole supported coefficient be priced using existing literature before enumerating its factor cells, and does the resulting separate-sign sieve estimate supply a positive twin margin? | ANSWERED | DERIVED from classical inputs: at delta=1/100 the harmonic limit is H_delta=D(25/12)-1+E_delta with 0<=E_delta<=24/22!, hence -13/2, where I is the explicit ordered two-factor integral below. A three-prime left input and rough composite right input suffice; prime partners are subtracted using an upper sieve. The witness survives every fixed delta<1/50 and every averaging profile with the same plateaus. Consequently the negative-only consumer is false, and liminf Phat/(C2*x)>1/2. The signed twin margin remains OPEN; literature novelty is unestablished. | [switching-negative-mass.md](switching-negative-mass.md) |\n| C | `Q-transition-energy-review` Does the sharp-energy chain in sharp-corner-transition.md survive an adversarial check, and do the upper norm, the small-fixed-eta lower norm and the failed negligible transition each stand at their stated scope? | ANSWERED | All ten assigned rows survive at their stated scope; no counterexample or decisive contradiction was found. The upper norm, the small-fixed-eta lower norm and the refuted negligible L2 transition each stand separately. Two repairs are citation-level only: the integer Barban-Vehov input is Graham 1978 quoted by Chen An, and the note should record why the k=Q constant is absolute. The de la Breteche-Dress-Tenenbaum PDF was refetched and its SHA-256 matches the pinned hash; its (1.5) and Theorem 1.1 supply exactly the imported (3) and (4). eta_0 remains positive and unextracted, and the signed sums, E_out and the twin margin remain OPEN. | [transition-energy-review.md](transition-energy-review.md) |\n| C | `Q-transition-energy-review-spec` What independent checks are required before the sharp-energy and non-negligible-transition conclusions can guide the next signed research attempt? | ANSWERED | Bounded review assignment with explicit source, algebra, interval, error and quantifier checks; per-claim verdicts and dependency consequences are required. The assignment is specified, not executed. | [transition-energy-review-spec.md](transition-energy-review-spec.md) |\n| C | `Q-transition-joint-budget` Can the mixed and smoothed terms be handled jointly with the transition pair, and what complete inequality would make such an estimate useful for the twin consumer? | PARTIAL | The exact split, cutoff-average identities and separate upper budgets survive. Averaging cutoff norms after triangle and Cauchy is saturated for sufficiently small fixed eta, but signed cutoff arguments and joint cancellation are not closed. R_00 need not be estimated separately under every grouping. The short-window constraint is corrected with a bounded majorant and a safely reduced lower constant. No consumer input or signed improvement is supplied; E_out and the sufficient twin margin remain OPEN. | [transition-joint-budget.md](transition-joint-budget.md) |\n| C | `Q-transition-joint-budget-spec` Can the mixed and smoothed terms be handled jointly with the transition pair, and what complete inequality would make such an estimate useful for the twin consumer? | ANSWERED | Bounded assignment to audit the exact four-term decomposition, price a concrete joint treatment, and return a sufficient one-sided or common-scale inequality with all open terms explicit. The decomposition and existing norm budgets are inputs, not new findings. | [transition-joint-budget-spec.md](transition-joint-budget-spec.md) |\n| C | `Q-transition-research-execution` How should the core research handler execute and integrate the three bounded transition assignments while retaining the central signed estimate? | ANSWERED | Prepared execution contract with a fixed mathematical baseline, three independently assignable specs, handler-owned signed research, dependency checks, shared-file ownership, reporting requirements and stopping rules. Execution is complete; the corrected results are in transition-round-review.md and transition-round-audit.md. The specification itself supplies no arithmetic estimate. | [TRANSITION-RESEARCH-EXECUTION.md](TRANSITION-RESEARCH-EXECUTION.md) |\n| C | `Q-transition-round-audit` Which conclusions of the completed transition round survive direct review, which require correction, and what research decisions remain justified? | ANSWERED | The sharp upper norms and sufficiently-small-fixed-eta lower norms survive. The integration contained substantive errors: an omitted lower cutoff, an incorrect Mellin formula for the transition, a conflation of singleton divisor fibers with long cofactor fibers, an unpriced source representation and overbroad closures of joint or cutoff arguments. Owning notes and current records are corrected. No signed improvement or sufficient twin margin is established; the corrected cofactor and joint-correlation interfaces remain open. | [transition-round-audit.md](transition-round-audit.md) |\n| C | `Q-transition-round-review` What survives the executed transition round after direct review, and what combined estimate is still missing? | ANSWERED | The energy chain survives at its original scope; the integration's missing cutoff, Mellin representation, fiber identification, normalization and general closure claims are corrected. The exact cofactor and joint-cutoff formulations remain useful. No signed estimate, controlled region, exact cut or sufficient twin margin changed. Only the stated norm-only procedure and the arbitrary-factor fixed-shift relaxation are refuted at their specified scopes. | [transition-round-review.md](transition-round-review.md) |\n| C | `Q-transition-signed-estimate` Can the signed transition kernel (E1) be estimated beyond its absolute norm budget, and if not, what exact input does the consumer need? | PARTIAL | No signed improvement. Derived: the proper-prime-power part of R_11 is O(x^(39/40+epsilon)); the prime part is an exact beta-weighted sum over pairs of linear forms of two-point Mobius correlations with coefficients up to x^(6/25+2eta) and x^(1/20+2eta); the Cauchy budget is at most 2B*eta*x*log^2(x)(1+o(1)) and the fiber triangle budget 4*eta^4*x*log^4(x)(1+o(1)), so every existing inequality loses one of the two signs. The cutoff-difference identity converts T at pn into a sharp-window coefficient times log(p)/L plus two boundary windows; the entropy-decrement route then needs bounded coefficients, a short-interval one-point input for T and a log-weighted consumer, none available. The required one-sided input is stated exactly in section 5. The twin margin remains OPEN. | [transition-signed-estimate.md](transition-signed-estimate.md) |\n| C | `Q-transition-source-match` Do the inspected divisor-correlation, shifted-convolution and averaged-Chowla statements directly supply a usable estimate for the exact transition kernel? | ANSWERED | No direct supplier was established among the inspected statements. The cofactor exchange and Theta identity are exact, but their scalar coefficient mass is not a full bounded-class representation for T. The transition Mellin kernel requires a sharp subtraction with a 1/s tail. Fixed-divisor singleton fibers differ from potentially long cofactor fibers. The fixed-shift arbitrary-factor relaxation is refuted; broader signed and cutoff methods remain open. | [transition-source-match.md](transition-source-match.md) |\n| C | `Q-transition-source-match-spec` What bounded source investigation could supply a usable signed estimate for the transition kernel, and how must its hypotheses and losses be priced? | ANSWERED | Specification for up to three detailed source-interface assessments, including coefficient representation, scale quantifiers, normalization, separation errors and exact downstream payoff. No new theorem match or literature-absence result is claimed. | [transition-source-match-spec.md](transition-source-match-spec.md) |\n| C | `Q-twin-reduction` What has the arithmetic campaign of 2026-09-05 and 2026-09-06 actually established about the twin sum, at what calibration, what exactly remains, and what would suffice? | ANSWERED | Consolidates the checked regional reduction S=C2*x+E_dagger+O_H(x/log^H x), its sufficient unbounded-scale margins, coefficient definitions and source dependencies. Later complete smooth and centered formulations are owned by RESEARCH-HANDOFF.md and their linked derivations. No new arithmetic estimate is supplied. Imported theorem proofs retain their stated review limits; every sufficient twin margin remains OPEN. | [TWIN-REDUCTION.md](TWIN-REDUCTION.md) |\n| 9 | `Q-applied-0828-varE` Were the Var/E red team's corrections applied to the four HELD notes? | ANSWERED | Every WEAKENED and REFUTED verdict in redteam-0828-varE.md is applied to its target note in the present tense; four ledger verdicts are rewritten, the theta=1 row is raised to PROVEN, and seven live-doc sentences plus one producer reading stay open and are drafted here. | [applied-0828-varE.md](history/staging/applied-0828-varE.md) |\n| 9 | `Q-lit-scourfield-2008` Does Scourfield, Smooth divisors of polynomials (LMS Lect. Note Ser. 352, CUP 2008, 286-311), supply the divisor-distribution estimate that varE-theta2-proof.md section 6 leaves open? | ANSWERED | DOES NOT DELIVER, at SECOND-HAND rung, and the chapter is still UNREAD at the page: ten channels were attempted, nine answered, and every one that answered is closed, degraded or paywalled, but the author's own restatement of the 2008 theorem, read at the page image of Scourfield, Funct. Approx. 55.1 (2016) 84 eq. (1.1), counts divisors m <= x of f(n) for n <= x with weight 1, for f a product of pairwise coprime irreducible factors each of degree at least 2, and the open step needs divisors n > 2L above the window, carrying a branch-dependent weight of total mass 4^omega, for the product of three LINEAR factors C(C-2)(C+2). Two gaps each independently fatal, plus a third that is AMBIGUOUS between two author-sourced statements and does not carry the verdict. One correction to lit-smooth-divisors.md is earned: the precision register of the owning convention is not uniformly (log y)^{o(1)} or order-of-magnitude, since (1.1) is an asymptotic with relative error O(1/log x), which is the right register and one epsilon short of the o(1/ln y) the open step needs. Step 1 of Var/E does NOT become conditional on a citation, and no absence may be written, since the chapter's interior lemmas were not read. | [lit-scourfield-2008.md](history/staging/lit-scourfield-2008.md) |\n| 9 | `Q-paper-iii-restructure` Is Paper III restructured around the proven model theorem? | ANSWERED | Yes. paper/variance-note.md now carries four new sections in place of its fit narrative: sec.8 the proven spectral form (Var/E = delta X, delta ln^2 W -> 16 C_2 e^{-2gamma}/3 = 1.109905, the Montgomery-Soundararajan main term exactly L prime by prime, so all of Var/E is discrepancy), sec.9 the decoupled model with Theorems 7 and 8 and Lemmas 9-11 written out with proofs and the closed form lambda_2(2) = 1 - e^{-2gamma}(9/2 - 4 ln 2) = 0.45546 plus the theta = 1 identity with Gorodetsky, sec.10 Conjecture 1 (delta(X - X_dec) -> 0) with eight exact ratios, two of three lag groups closed and the third named as a divisor-distribution statement not yet searched, and sec.11 the 0.611 refutation as a short calibration paragraph; sections 1-5 are untouched, the abstract and section 6's open question are rewritten to match, and the \"u = 2 exactly\" defect in section 7 is corrected. | [paper-iii-restructure.md](history/staging/paper-iii-restructure.md) |\n| 9 | `Q-recon-0830-smooth-aps` Does any theorem on y-friable integers in arithmetic progressions (Granville, Fouvry-Tenenbaum, Soundararajan, Harper, Drappeau, Drappeau-Granville-Shao) supply the uniform o(1) equidistribution that attack-0830-varE-identification.md section 4 leaves open, for moduli up to y^(4/5) with the lam1 weight and the squarefree restriction? | PARTIAL | NONE APPLIES AS STATED, and the step stays open; lim Var/E = 0.45546 stays HEURISTIC. Read at the page (eleven sources, arXiv text layers and the Acta Math. page image): every pointwise asymptotic is O(log q / log y)-precise (Granville 1993 Thm 1) or hypothesises log x / log q -> infinity, and every beyond-square-root level (Fouvry-Tenenbaum 1996, Drappeau 2015, Drappeau-Granville-Shao 2017, Pascadi 2023/2025) hypothesises y <= x^delta, i.e. u -> infinity, against our u in (1.2, 2]. NEAREST: Harper arXiv:1208.5992 (2012, a preprint with no journal version found) Theorem 2, the Barban-Davenport-Halberstam form, whose range (log^K x <= y <= x, Q <= Psi/log^A x, bounded u in the ineffective form) covers d up to L^(1/2)/log and whose weight (1 on all friable n, no squarefree restriction, no max over x' <= x) is unmet. The record's quantifier is also refined: the cell needs the lam0-weighted AVERAGE over d, not uniformity. PROVEN given the weighted form (H_w) of that theorem, derivation mine and unreviewed, with its two identities checked numerically in the producer: every two-branch cell of Xmix, both mirror cases, all bands, balanced blocks included, is o(ln^2 y), so the Bettin-Chandee kernel separation is not on this route; the reduction of (H_w) to Harper's theorem is OUTLINED and not written, and the three-branch type (measured -0.0002 at x = 19, trivial bound O(ln^3 y)) is untouched. Not TPC-strength. | [recon-0830-smooth-aps.md](history/staging/recon-0830-smooth-aps.md) |\n| 9 | `Q-redteam-0828-varE` Does the 2026-08-28 chain from the exact comb variance to lim Var/E = Pr[GD(2) > 2] = 0.45546 survive an adversarial re-derivation, and does the refutation of the 0.611 reading hold? | ANSWERED | The constant survives at HEURISTIC and the 0.611 refutation is CONFIRMED and strengthened (the frozen out-of-sample half of the protocol also fails on the control); the correction is that TWO steps are open, not one, and that the theta=1 branch is an exact identity with a read theorem rather than a second-hand numerical match. | [redteam-0828-varE.md](history/staging/redteam-0828-varE.md) |\n| 9 | `Q-redteam-0830-imports` Do the six lighter notes of 2026-08-30 (varE-identification, smooth-aps, rec-killrun, skeleton-door, at43-bigint, coherence) survive an adversarial pass on independent code and at the source pages, and which of their load-bearing claims are wrong? | ANSWERED | No note is refuted whole and no exponent moves. Every decisive measurement reproduces on code sharing nothing with the producers: the varE mixed remainder and its 0.844 / 0.826 share below 2L and 94.4 / 98.7 percent unbalanced share at x = 13, 17; the Buchstab deviations and all three crossing levels log10 z* = 33.53, 68.80, 33.62; the skeleton door's 9.2 / -0.9 percent, G30_agg 0.1113 / 0.1011 and the 0.0102 cap at @13, @17, by a path that also verifies cap-36 Theorem A; the @43 parity exactness on all 26,157,448 scour classes with 0 above 2^54 and a constructed @47 counterexample; and the five recon-0828 verdict counts summing to 75, which REFUTES rec-killrun's brief-error clause and confirms the orchestrator's rider. Seven claims WEAKEN. The load-bearing one is smooth-aps's \"the weight is unmet in print\": Harper's J. London Math. Soc. 112 (2025) e70293 proves a Barban-Davenport-Halberstam asymptotic for an ARBITRARY sequence on exactly that dyadic all-classes object, and his own footnote says the missing max over x' can be incorporated, so the NEAREST citation changes even though the step still does not close (the 2025 range is sqrt(2x) < Q <= x and its Theorem 2 route bars sieved sets by name). Also: Henriot Corollary 2 applies, but to the ERRATUM's statement and under two hypotheses the note leaves unstated (eps < alpha/600, not \"every eps\"; F at prime powers); varE section 4's display (*) carries an unfilled placeholder and is vacuous by a factor L/d; the skeleton door's \"would tighten\" is wrong in sign at @17, where the closable block is negative and removing it RAISES G30_agg to 0.1020; Halasz-Montgomery is not applicable for the reason given (it is an abstract inner-product inequality); FGKT's heuristic quote is on p. 4, not p. 3; and the \"~13x had no source\" is a corpus miss, the retired G2-walk figure at CHANGELOG.md:5716. Not TPC-strength; nothing here opens a route. | [redteam-0830-imports.md](history/staging/redteam-0830-imports.md) |\n| 9 | `Q-redteam-0830-slack` Do attack-0830-head-remainder.md, attack-0830-tail-derivation.md and verify-0830-record-defects.md survive an adversarial re-derivation on independent code, and do the five riders the verification put on live notes stand as written? | ANSWERED | Arithmetically they survive: every figure of the three notes that this pass could recompute reproduced to the printed digit on code written from the definitions, 0 assertion failures over 58 assertion call sites (most inside per-level loops), including every figure of the verification's three claims and the half-decade Delta_HL it quoted rather than recomputed. Four sentences do not survive. (1) verify-0830-record-defects.md:316, a replacement queued for head-residual-hl3.md sec.0, says the pooled value sits \"0.038 to 0.052 below the sub-window mean at all three decades\"; measured, the two lower decades run to 0.076 and 0.066, and the same note's own falsifier row says 0.038 to 0.076, so the wrong half is the one queued to land. (2) verify-0830-record-defects.md:379's un-measured \"+0.005\" residual inside the eighths is +0.0004 measured at sixteenths, so the de-pooled top-decade value is converged and the caution can be dropped. (3) the ruling that the class null is \"the matched figure\" for the tail is matched on residue class only: p'^2 is coprime to every q <= p', and the ensemble's rough-class offset is E_rc - R = 5.6714 -> 7.5357 over x = 7..29 against the class null's 5.6000 -> 6.0336, giving t/(R_shell + 7.5357) = 1.0123 against t/classNull = 1.0157. (4) attack-0830-head-remainder.md sec.3's top row is not gap-scale matched: E[g] 244.0 against 235.9, y_match capped at 19997, and on the note's own six ensemble points Delta_ens ~ E[g]^-0.312, so meas/ens reads 0.9640 not 0.9539 and the headline \"0.954 to 0.995\" reads 0.964 to 0.995. Three levels the notes recorded as out of reach are run here: the X2 group at x = 23, the tile identities and conditionals at x = 29, and the ensemble ladder at y = 29 (pipe/ens 0.9535, 40 s, against the head note's \"about fifteen minutes\"). No route opens or closes and nothing here touches Z2. | [redteam-0830-slack.md](history/staging/redteam-0830-slack.md) |\n| 9 | `Q-var41` What does the stable law predict for Var(41), and what can the tenth Var/E point pin? | OPEN | Pre-registration only, sealed and committed alone before any Var(41) engine exists: it freezes the prediction, a band taken from the law's own residuals at z <= 37, the derived z(41) prediction, and the honest statement that one more point cannot separate a limit from a drift. | [var41-prereg.md](history/staging/var41-prereg.md) |\n| 9 | `Q-varE-identification-0830` Can either open step behind lim Var/E = 0.45546 (the identification delta*(X - X_dec) -> 0, or the theta = 2 mean-coefficient replacement) be proven, and if not, which single inequality does not close? | PARTIAL | Neither closes, and the two named steps are one statement (Conjecture 1 of variance-note sec.10; the model half is PROVEN in varE-limit-theorem.md). The remainder is re-split exactly at five levels: the corpus's X2 column doubles the positive half of one shift pattern (X2c = 2.66 against the group sum 0.09 at x = 19, PROVEN by rebuild), so its Xmix is overstated 2.3x; the true mixed remainder is MEASURED at -0.49 ln y and 82 % of it sits BELOW 2L, in the two-branch type, in the unbalanced range min(d,e) <= L^(2/5) (99.7 % of that cell at x = 19), outside the range of every bilinear Kloosterman-fraction bound; the divisor-distribution estimate the record names (n > 2L only) is therefore not the whole open step. Henriot's Corollary 2 (read at the page, erratum read) applies as stated and cuts the moduli to n <= L ln^(2+o(1)) y, PROVEN given that theorem; the one inequality left is the uniform o(1) equidistribution of y-friable squarefree integers in progressions to moduli up to y^(4/5), NOT SEARCHED at the page. The limit 0.45546 stays HEURISTIC. Not TPC-strength. | [attack-0830-varE-identification.md](history/staging/attack-0830-varE-identification.md) |\n| 9 | `Q-varE-limit` Does lim Var/E on the diagonal window exist, and what is it? | PARTIAL | The theta=2 mean-coefficient step is neither proven nor refuted: the replacement error is exactly a sum over shifts of W(h) - V(h), it splits into a c=0 group (needs no decoupling) and CRT-mixed lags (open); measured ratios true/model 1.0098, 1.0039, 1.0014, 1.0013 at x = 13..23 with delta(X - X_dec) ln y falling rather than settling, so the error is O(1/ln y) or better MEASURED on four levels that exclude growth and nothing finer, and the limit 0.45546 stays HEURISTIC with varE-spectral's second step, its own limit theorem, still open. | [lit-dickman-variance.md](history/staging/lit-dickman-variance.md), [lit-smooth-divisors.md](history/staging/lit-smooth-divisors.md), [varE-asymptotic.md](history/staging/varE-asymptotic.md), [varE-exact-ladder-01.md](history/staging/varE-exact-ladder-01.md), [varE-limit-theorem.md](history/staging/varE-limit-theorem.md), [varE-spectral.md](history/staging/varE-spectral.md), [varE-theta2-proof.md](history/staging/varE-theta2-proof.md), [varE-theta2-step.md](history/staging/varE-theta2-step.md) |\n| 9 | `Q-verify-record-defects-0830` Do the three defect claims of 2026-08-30 (attack-0830-varE-identification.md sec.8 on the doubled X2 column; attack-0830-tail-derivation.md sec.7 on the corpus's \"R + 1/2\"; attack-0830-head-remainder.md sec.1 on the decade-pooled Delta = 0.6214) reproduce on independent code from the definitions, and which of the corrections they list apply? | ANSWERED | Claim 1 CONFIRMED: the corpus delta*X2 column is the positive half of the p \\| h-2 pattern doubled, that pattern's mirror is the p \\| h+2 pattern (W-(-h) = W+(h) exactly, W-(h) != W-(-h) at 9 to 900,679 shifts), the group sum 2X2 reads -0.5603 .. 0.1846 at x = 7..19 against 5.7147 .. 5.3266, and Xmix is overstated 2.305x at x = 19 (6.9x at x = 7); X, X1 untouched. Claim 2 AMENDED: R + 5/2 and R + 3 are exact for the tail convention (all and odd origins, asserted at x = 7..23) and t/(R + 3) = 1.0228, but head-residual-factor.md:70 is the HEAD's forward convention where R + 1/2 and R + 1 are exact, so that correction does not apply; and p'^2 is 1 mod 6, where the tail constant is 5 (t/(R + 5) = 1.0181), with the class null (1.0157) the matched figure, so 1.0228 is one unmatched convention replacing another. Claim 3 CONFIRMED as an artefact, AMENDED on mechanism and on the half-decade reading: 0.6214 pooled against 0.6592, 0.6705, 0.6736 at 2, 4, 8 sub-windows; the between-slope (Simpson) piece the note derives is half of the pooling term (0.0187 of 0.0378 at halves, 0.0261 of 0.0522 at eighths), the other half is the within slope's own fall with height weighted by Var(g); and re-reading the halves from their eighths puts HL BELOW the measurement at 5 of 6 half-decades by 2.5 to 3.4 percent at the top decade, so \"within 1.6 s.e., sign alternating\" is itself a pooled statement. | [verify-0830-record-defects.md](history/staging/verify-0830-record-defects.md) |\n| Z2 | `Q-X-upper-0829n` Can the rough-pair census X(K) on the stretch window be bounded above, unconditionally and with an explicit error term, by a dimension-2 upper-bound sieve, and how loose is the bound against exact X(K)? | PARTIAL | Yes, as a theorem with nothing open (Selberg's Λ² inequality with a remainder bounded by the interval structure alone, VERIFIED at all 10908 anchor-depth pairs), but it is loose by 7.82 to 10.13 against exact X(K) at the band depths, decomposed as a sieve loss 1.78 to 2.53 times a structural loss 4.01 to 4.39 that no upper-bound sieve can remove, the window holds the level below z² (s_max 1.81 to 1.96) so the bound stops improving past p_K = 31 to 61, its main term is the sifting function's and not the census's, and the drift is a crossing quantity outside the legal half and is not evaluated. | [attack-0829n-X-upper.md](history/staging/attack-0829n-X-upper.md) |\n| Z2 | `Q-applied-0828-quadpoint` Were the quadpoint red team's corrections applied to the four HELD notes? | ANSWERED | Applied: the nine REFUTED and WEAKENED verdicts that target the four notes are written into them in present tense, with ledger blocks added; the tenth targets the sealed prereg, outside the fence, and is recorded in attack-quadpoint-02.md instead; three corrections owed by TODO.md Z2 and one by Z0 are listed here and are NOT applied. | [applied-0828-quadpoint.md](history/staging/applied-0828-quadpoint.md) |\n| Z2 | `Q-capture-identity` Does the transplanted cap family collapse onto a classical object, and what does the collapse say about the depth law and the wall? | ANSWERED | PROVEN, elementary: floor_K = T − X(K) on the half-open window with both members in [Q², Q′²), a convention that is load-bearing; the wall is renamed, not weakened; Σ capU counts each member against its own threshold, not against a uniform p_K, the two differing 11% to 37% at K*. | [quadpoint-identity-01.md](history/staging/quadpoint-identity-01.md) |\n| Z2 | `Q-derive-0904-r0-extension` Which pieces of the R0 decomposition (the tile form, the zone form, head, tail, Z2, and the composites R0/R1/R2/stretch) carry the parity obstruction's forbidden extension, and is any single piece both legal on the TPC-strength axis and exempt from it? | ANSWERED | No piece of R0 is both legal and exempt with content: the criterion reduces to a coordinate count (Proposition A) under which every statement forcing a twin pair into a range where roughness implies primality is covered, and the exempt residue is exactly the statements that assert no such forcing (R0's identity, lower bounds, boundary artefacts) plus the tile form, whose exemption is a property of the statement and provably does not close under deduction (Proposition B), so it survives only for routes with no sieve weight anywhere. | [derive-0904-r0-extension.md](history/staging/derive-0904-r0-extension.md) |\n| Z2 | `Q-parity-adversary` Does a parity adversary built on our own stretches survive at every level, and what precision class does a certificate need? | ANSWERED | It does NOT survive: the exact-data adversary dies at D* about Q^1.18 (MEASURED, exact arithmetic, 43 anchors Q <= 200), and the certificate that kills it is itself destroyed by 0.30 of one count per modulus, so Z2's precision class is now measured rather than described; it is PROVEN not to be a proof ingredient. | [attack-parity-adversary.md](history/staging/attack-parity-adversary.md) |\n| Z2 | `Q-parity-dstar-law-0829n` Does the exact-data parity adversary's killing level D*(Q) follow D* ~ Q^c with a stable c on the decades past Q = 200, and is c distinguishable from the independent-thinning null's? | CLOSED | No law in Q and CLOSED by the sealed kill rule (score 5 HIT, 1 MISS): MEASURED on 271 anchors 211 <= Q <= 13679, the Q-slope moves 1.446 to 1.204 to 0.575 across ranges while ln D* on ln width holds slope 1.097 +- 0.012 with a quarter of the scatter, D*/width about 0.5, and the independent-thinning null reproduces the width law within 0.025 in slope; the mechanism is that D* is where a modulus first reads a single position of the window, which is the wall survey's remainder statement in the stretch coordinate and names nothing new. | [attack-0829n-parity-dstar.md](history/staging/attack-0829n-parity-dstar.md) |\n| Z2 | `Q-quadpoint-blind-31607-0830` Does the zero-parameter depth law y*(Q), stated on the set where T(Q) >= 1 and on both the 1/(2e^gamma) = 0.280730 and the u*omega(u) = 2 (1/u* = 0.280438) forms with and without the finite-size Mertens factor, survive a sealed blind test on the bands past Q = 10007, and can that decade tell the two asymptotes apart? | ANSWERED | MEASURED, blind, and it decides nothing asymptotic: on 8363 new anchors 10009 <= Q <= 100003 the 76 sealed rows score 64 HIT and 12 MISS with every MISS on the two naked asymptotes as pre-declared, the two finite-size zero-parameter comparators HIT every row at every band (residual on ln y*/ln h -0.00022 to -0.00002 against C2), K*/pool falls 0.032 -> 0.026 -> 0.020 -> 0.016 -> 0.013 with no bend and a quiet sup, and the band means sit 0.0017 to 0.0037 below both asymptotes, six to twelve times their 0.000291 difference, so which limit the deficit tends to is set by the comparator and not by the data. | [blind-0830-quadpoint-31607.md](history/staging/blind-0830-quadpoint-31607.md) |\n| Z2 | `Q-quadpoint-c1c2-0830` Does the fourth decade's post-hoc, unsealed preference for the omega-form comparator C2 over the exact-Mertens comparator C1 (C1 above the data by 3.46 and 8.83 se at B11 and B12, C2 within 2.82 se) survive a sealed blind test at the fifth decade, Q in (100003, 316243], and what does that say about which mechanism explains the depth law's drift? | ANSWERED | MEASURED, sealed before the run and scored blind on 17702 new anchors, and it decides nothing asymptotic: the sealed D1 row HITs at both B13 and B14 (C1 above the data by 13.33 and 23.82 se, C2 within 1.26 and 1.82 se), so by the sealed rule the omega-at-finite-u crossing of import-rough-anatomy.md §2.1 is CONFIRMED over the exact-Mertens product of u2-engine-depth.md §5 as the depth law's main term at the sampling scale, while both still HIT every ±0.0015 row, the kill rows on C2 all HIT, K*/pool falls 0.013 -> 0.010 -> 0.008 with no bend, and the band means sit 0.0012 to 0.0018 below both asymptotes, so the two constants 0.280438 and 0.280730 stay unseparated by the pre-declared rule and the run separates two mechanisms, not two limits. | [blind-0830-quadpoint-c1c2.md](history/staging/blind-0830-quadpoint-c1c2.md) |\n| Z2 | `Q-quadpoint-prior-art` In the owning conventions, who holds the rough-pair census, the capture identity and the crossing law, and is any of it ours? | ANSWERED | The object and the identity are standard (the dimension-2 twin sifting function plus Buchstab's identity; ADJACENT-STANDARD, no verbatim carrier, no novelty language); only the comparison X(y) < T and the crossing law stay ABSENT-PER-CONVENTION, and the Buchstab correction is 6.6% of the top-band residual, not 2%. | [quadpoint-prior-art.md](history/staging/quadpoint-prior-art.md) |\n| Z2 | `Q-quadpoint-transplant` Does the anchored-cap family transplanted onto the stretch window S_Q certify every anchor, and what does the depth cost do with height? | ANSWERED | Pre-registration for the decade extension, committed ALONE before the segmented producer existed: it fixes the FALL trend line, the registered band readings, the kill rule and a digit-exact calibration gate that must abort before any new figure prints, and states that a falling K*/pool over two decades would still be only a measurement. | [attack-quadpoint-01.md](history/staging/attack-quadpoint-01.md), [attack-quadpoint-02.md](history/staging/attack-quadpoint-02.md), [quadpoint-decade-prereg.md](history/staging/quadpoint-decade-prereg.md) |\n| Z2 | `Q-recon-0828-rough` Does the 2000-2026 literature on y-rough numbers, sifted sets and short intervals hold any theorem that lowers Z2 or G2, meaning any control on the largest gap or the minimum count of a sifted set in EVERY short interval, or any structure of rough numbers a gap bound could use? | ANSWERED | No. Ten angles examined, ten killed for this purpose; the strongest every-interval statement in print for one-class rough numbers is Iwaniec's S(x,y,z) >= (4y/log^2 y)(log(y/z^2) - O(1)) for y >> z^2, uniform in x, giving J(P(z)) << z^2, and it is the linear sieve run exactly at its sifting limit beta_1 = 2 with no slack outside it; no two-class analogue exists in print at any exponent, the corpus's own x^4.2665 is beta_2 by the same mechanism, and every remaining branch of the field (Matomaki-Radziwill, Gorodetsky, Friedlander-Iwaniec Cor 6.28, Gafni-Tao) carries the almost-all quantifier, which a second-moment statement can never convert into a bound on one empty window. | [recon-0828-rough.md](history/staging/recon-0828-rough.md) |\n| Z2 | `Q-recon-0828-sieve` Is there anything in the 2008-2026 sieve literature, or in the parity-breaking literature, that lowers beta_2 = 4.26645 or supplies a consumer for a level theta > 1 on the two-class comb? | ANSWERED | No route. Twenty-one angles examined, nineteen dead on arrival, nothing in print beats 4.26645 at kappa = 2 at any level on four calibrated channels; the two survivors are an owning-convention row and one unresolved hypothesis in a five-page 1990 announcement whose stated requirement beta < 3 excludes beta_2 = 4.26645. | [recon-0828-sieve.md](history/staging/recon-0828-sieve.md) |\n| Z2 | `Q-redteam-0828-quadpoint` Do the four HELD quadpoint notes of 2026-08-22 survive an adversarial pass: the capture identity's proof line by line, the depth law and its asymptote, the sealed prereg scoring, and the prior-art citations at page level? | ANSWERED | The identity is PROVEN once its window convention is written down and holds at every K of 50 anchors under an independent implementation; four load-bearing sentences are wrong (section 4's gloss on Sum capU, v1 section 2's limit, the prior-art note's 2%, and the note's unconditional y* sentences), and none of the four defects opens or closes a route. | [redteam-0828-quadpoint.md](history/staging/redteam-0828-quadpoint.md) |\n| Z2 | `Q-redteam-0830-records` Do attack-0829n-parity-dstar.md (CLOSED, HELD) and attack-0830-record-mechanism.md (PARTIAL, HELD) survive an adversarial pass on independent code, does REFUTED row 95 stand as worded, and is TODO Z5's new title supported? | ANSWERED | Both notes' measurements reproduce on independent code and neither headline is refuted: D* is confirmed EXACTLY, by a BigInt rational simplex written here, at 299 of 376 anchors with 0 contradictions (the other 77 exceed the exact solver's size cap), the old range's 172-figure gate is re-derived exactly at all 43 anchors with 0 disagreements, and an independent 30-wheel sieve to 1e11 reproduces note (b)'s seven decade gap counts, CV^2 and far-tail slopes to every printed digit, with the counts also matching the published pi_2(10^k). Four sentences are wrong or over-stated and are replaced here: REFUTED row 95, note (a)'s ledger, its section 6 and TODO Z5 all attribute the null's 0.025 agreement to the WIDTH law when 0.025 is the ln Q slope difference (the width-slope difference on the same 84 matched anchors is 0.031); note (a) section 3's \"the threshold sits in a gap of at least 0.25 at every anchor\" is scoped in its own producer to the true arm and the null arm's exact gap is 0.1212; P6's HIT rests entirely on the decade-2 collapse that section 5 discounts as a sampler artefact when it discounts kill clause (a); and note (b)'s \"d b_z = 1.0806 +- 0.0014\" carries only the Monte Carlo pairing error, while moving the far-tail slope by 0.02 (against a spread of 0.0174 across six fit windows on the same decade and 0.1196 across decades) moves d b_z from 0.9888 to 1.1720, and continuing the law at slope 1 above the last resolved point still gives 0.7752. REFUTED row 95's CLOSED verdict stands on kill clause (b), which is better powered than the note presents it (paired s.e. 0.0401 against a 0.15 band), and the row needs two clauses reworded; TODO Z5's title is supported for \"measured and not derived\", and the resolved part of the law carries 72 per cent of the residual on its own with the extrapolation carrying the rest. | [redteam-0830-records.md](history/staging/redteam-0830-records.md) |\n| Z2 | `Q-redteam-0830-zone` Do attack-0829n-X-upper.md, blind-0830-quadpoint-31607.md and blind-0830-quadpoint-c1c2.md survive an adversarial re-derivation and an independent re-implementation of their producers, and at what rung does the corpus carry each of their headline claims? | ANSWERED | They survive on the numbers and mostly on the wording. On independent code every band sum, tier ratio, sealed forecast, band statistic and z of all three notes reproduces to the printed digit (X-upper 7.82 to 10.13 and the flat 378100.0; the fourth decade's 76 rows at 64 HIT and 12 MISS with the same twelve MISSes; the fifth decade's class A at both bands), and three attacks failed to break them (the paired-se attack on the D rows, a search for a fitted constant in C1, and gap-selection at the band edges). Four defects stand: the X-upper note's \"s_max 1.81 to 1.96\" is wrong in its own arrow convention, which means B3 to B8, since s_max peaks at B4 and reads 1.85 at B8; the c1c2 note's OLD R2 row at B14 is a boundary call that my 1-in-8 subsample scores the other way; the c1c2 note's custody line \"three tails, one code-sha256\" is a tautology of tailfmt.js and is not evidence of a seal; and the unscored y* check carries a Jensen bias of about +0.44 in y, which is the whole size of the C2 gap it is read as corroborating. Nothing here moves 4.2665 to 2, no rung rises, and all three notes stay HELD. | [redteam-0830-zone.md](history/staging/redteam-0830-zone.md) |\n| Z2 | `Q-redteam-0904-r0-extension` Do the seven load-bearing claims of derive-0904-r0-extension.md survive an adversarial re-derivation against Tao's own text, and do its proposed live-layer changes survive? | ANSWERED | Six of seven survive at their stated rung and one is refuted: Proposition A, the scope slip, row e'' and the equivariance transfer are CONFIRMED (the slip is strengthened, since Tao's own Example 2 is the target note's missing one-coordinate witness), Propositions B and D are WEAKENED (B's stated mechanism proves too much and its proposed insertion drops the qualifier that saves it; D's \"same trigger\" is near-definitional and its causal reading of the exponent band does not hold), and the claim that no corpus artifact certifies the window-count variance at width p'^2 is REFUTED by research/06-variance-theorem.js, which computes exactly that, brute-force verified at two levels; separately, EXEMPT throughout means \"Claim 1's H5 fails\", which is not the parity statement wall-note Face 1 is priced against, and the note's proposed live-layer text carries no H6 conditional. | [redteam-0904-r0-extension.md](history/staging/redteam-0904-r0-extension.md) |\n| Z2 | `Q-roughpair-error` How big is the empirical error of the rough-pair census, and is Z2 dead on the numbers? | PARTIAL | Not dead on the numbers - the error sits below T at every one of 1,206 anchors, its scatter is sub-Poisson, its systematic part is 0.1% at the top band with the favourable sign and its exponent is 0.40 below the twin count's - but the same run moves the wall off the error term onto the depth, where the slack at the crossing is 1.056 and falling and the main term alone is 5.6 times the twin count at the sifting limit. | [attack-roughpair-error.md](history/staging/attack-roughpair-error.md) |\n| Z2 | `Q-roughpair-null` Does the independent-thinning null named in attack-roughpair-error.md's defects list return chi2/df = 1, and is the measured sub-Poisson dispersion of the rough-pair census error real or a normalisation artefact? | ANSWERED | MEASURED, null-side, label (i), no bearing on the certificate: the independent-thinning null is binomial and returns 1 - p (0.8164 to 0.9808), not 1, so attack-roughpair-error.md's \"below 1 everywhere: sub-Poisson\" is WEAKENED and the registered F1 fires at u = 5; the deficit against the mean-matched null survives at pooled z = -5.87, -7.09, -11.39; and a CRT-exact null that only makes each prime's kill count exact in the window predicts 0.33 to 0.56, which the measured value EXCEEDS by 1.26 to 1.40, so D8 is interval-level in form, elementary in content, and not wall-relevant. | [measure-roughpair-null-0829.md](history/staging/measure-roughpair-null-0829.md) |\n| Z4 | `Q-applied-0828-census` Were the census red team's corrections applied to the three HELD notes? | ANSWERED | Every WEAKENED and REFUTED verdict in redteam-0828-census.md is applied to its target note in the present tense; two producer defects and four live-doc corrections stay open and are listed here. | [applied-0828-census.md](history/staging/applied-0828-census.md) |\n| Z4 | `Q-applied-0828-head` Were the head red team's corrections applied to the three HELD notes? | ANSWERED | Applied. Every WEAKENED and REFUTED row of redteam-0828-head.md lands in the note it targets as 42 prose edits, and all three ledger verdict lines now carry the surviving chain; four corrections fall outside the fence and are owed elsewhere, the load-bearing one being destroyer-census-01 §6(b)'s R and its h/R = 1.09 -> 1.03. | [applied-0828-head.md](history/staging/applied-0828-head.md) |\n| Z4 | `Q-applied-0828-live` Which engine and head red-team corrections reached the live layer? | ANSWERED | Nineteen edits across four documents: the glossary's cap_K dimension count with its mandatory sifted-versus-assumed clause plus three new entries (twin opener, head, tail); the survey's one-class density factor, S1's restored pi(q-1) terms and r in N_x hypothesis, the dropped Chen absorption, Theorem C with its emptiness, dimension 1 for q_i < q, the comb-conditioning qualifier and the inverted shallow-band bullet; the certificate engine's equidistribution status cell, the q = T^{o(1)} second hypothesis and the Comb Discrepancy scope cell; and the census's R as the continuum functional with its three comparators; paper/staircase-note.md, TODO.md and QUESTIONS.md are owed and untouched. | [applied-0828-live.md](history/staging/applied-0828-live.md) |\n| Z4 | `Q-destroyer-census` Who destroys each channel pair, what does the zone kill budget certify, and where does the counting certificate die? | ANSWERED | The census reproduces digit-exactly on an independent engine; the destroyer rule is positional, not temporal, the certificate holds at 15 zones and dies at p = 67 where B/C first crosses 1, and no first-mover excess survives the derived null. | [destroyer-census-01.md](history/staging/destroyer-census-01.md) |\n| Z4 | `Q-head-remainder-0830` Does the 5.3 percent remainder that Hardy-Littlewood leaves in the head's endpoint deficit Delta = 2 - beta derive by a route that is not HL: the frozen-sieve first-survivor object, a level-of-distribution correction, or the exact tile-ensemble head? | PARTIAL | No route that is not HL closes it, and the remainder as recorded is not what it was taken to be. The record's Delta_meas = 0.6214 at [1e7,1e8) pools a decade over which the prime density falls ten percent, and the OLS intercept absorbs the gradient (a Simpson-type effect, sign derived): the two half-decades read 0.6848 and 0.6500, the four quarter-decades 0.7203 to 0.6565, and pooled minus the gap-weighted mean of the halves is -0.0378, -0.0455, -0.0524 at the three decades, so the same-sign 5.3 percent miss is the pooling. At half-decade resolution HL's pipeline prices Delta to 1.2 and 0.4 percent at the top two windows, inside one standard error, and the sign of the miss alternates. That agreement is a near-cancellation of two four-percent effects: the pipeline under-predicts its own rotation ensemble by 3.6 to 4.9 percent (pipe/ens 0.951 to 0.964 across y = 313 to 19997, MEASURED by exact CRT sampling; 1.049 to 0.958 at y = 11 to 23 by full enumeration), an ensemble quantity that derives by finite count with no prime input and whose sign is opposite to hl3 section 2's guess; and the anchored window's deficit sits 2 to 5 percent below its gap-scale-matched ensemble (Delta_meas/Delta_ens 0.878 to 0.995 over six half-decades, 0.954 to 0.995 on the four with more than 13,000 gaps), the anchoring part, which is a three-point prime correlation in a short window and is HL-strength by every route named. The head is a slack field and nothing here touches Z2. | [attack-0830-head-remainder.md](history/staging/attack-0830-head-remainder.md) |\n| Z4 | `Q-head-residual` What is the head's residual prime-origin factor h/R = 1.09 -> 1.03, and does it derive? | PARTIAL | h - R decomposes exactly and A_forced derives; h/R -> 1 follows from beta CV^2 being bounded, beta CV^2 -> 2 controlling the RATE and not the limit (under it h - R -> D = -0.95, so +8.4661/ln^2 p cannot be the limit law); R is the continuum functional and the integer-origin null sees R + 1/2 exactly, so h - R = 5.679 depends on the unstated null population, reading 5.179 discrete-uniform and 2.925 coprime-to-30; the mod-30 answer is NO by a valid route, class term -0.0001; Delta = 2 - beta is priced by HL to 5.3 percent with nothing fitted and HL's own 1/ln x term absorbs the miss, so the 5.5 s.e. framing is dropped; the asymmetry-as-mechanism claim and the [0.289, 0.361] bracket are refuted, hl3's 16 percent standing; beta_null is 1.898 for the discrete-consistent null and 2 in the continuum; the 72/28 split reads 81/19 in the other exact order; c = 0.844 is 0.722 on a subset, pinned to 15 percent; Delta_HL ln p/lnln p = 4.02 is a property of the prediction; and the N4 plateau is not OPEN, needing beta CV^2 = 2.71 rising to 6.63. | [head-residual-factor.md](history/staging/head-residual-factor.md), [head-residual-hl3.md](history/staging/head-residual-hl3.md), [head-residual-null.md](history/staging/head-residual-null.md) |\n| Z4 | `Q-redteam-0828-census` Do destroyer-census-01.md, stretch-01.md and records-placement-01.md survive an independent re-derivation of every load-bearing claim, and may they leave HELD? | ANSWERED | All three survive on arithmetic, every decisive number reproducing exactly on an independent engine; one sealed reading (records READ-3) is REFUTED as vacuous and four sentences need correcting before integration. | [redteam-0828-census.md](history/staging/redteam-0828-census.md) |\n| Z4 | `Q-redteam-0828-head` Do the three 2026-08-28 head-residual notes survive an adversarial re-derivation of every load-bearing claim on independent code? | ANSWERED | Mostly yes on the numbers and no on two framings: \"h/R -> 1 iff beta CV^2 -> 2\" is REFUTED (the limit needs only boundedness, the product controls the RATE), and 9 to 59 percent of the quoted h - R is the choice of comparator, since R is the continuum functional and a discrete uniform origin sees R + 1/2 exactly. | [redteam-0828-head.md](history/staging/redteam-0828-head.md) |\n| Z4 | `Q-redteam-0830-slack` Do attack-0830-head-remainder.md, attack-0830-tail-derivation.md and verify-0830-record-defects.md survive an adversarial re-derivation on independent code, and do the five riders the verification put on live notes stand as written? | ANSWERED | Arithmetically they survive: every figure of the three notes that this pass could recompute reproduced to the printed digit on code written from the definitions, 0 assertion failures over 58 assertion call sites (most inside per-level loops), including every figure of the verification's three claims and the half-decade Delta_HL it quoted rather than recomputed. Four sentences do not survive. (1) verify-0830-record-defects.md:316, a replacement queued for head-residual-hl3.md sec.0, says the pooled value sits \"0.038 to 0.052 below the sub-window mean at all three decades\"; measured, the two lower decades run to 0.076 and 0.066, and the same note's own falsifier row says 0.038 to 0.076, so the wrong half is the one queued to land. (2) verify-0830-record-defects.md:379's un-measured \"+0.005\" residual inside the eighths is +0.0004 measured at sixteenths, so the de-pooled top-decade value is converged and the caution can be dropped. (3) the ruling that the class null is \"the matched figure\" for the tail is matched on residue class only: p'^2 is coprime to every q <= p', and the ensemble's rough-class offset is E_rc - R = 5.6714 -> 7.5357 over x = 7..29 against the class null's 5.6000 -> 6.0336, giving t/(R_shell + 7.5357) = 1.0123 against t/classNull = 1.0157. (4) attack-0830-head-remainder.md sec.3's top row is not gap-scale matched: E[g] 244.0 against 235.9, y_match capped at 19997, and on the note's own six ensemble points Delta_ens ~ E[g]^-0.312, so meas/ens reads 0.9640 not 0.9539 and the headline \"0.954 to 0.995\" reads 0.964 to 0.995. Three levels the notes recorded as out of reach are run here: the X2 group at x = 23, the tile identities and conditionals at x = 29, and the ensemble ladder at y = 29 (pipe/ens 0.9535, 40 s, against the head note's \"about fifteen minutes\"). No route opens or closes and nothing here touches Z2. | [redteam-0830-slack.md](history/staging/redteam-0830-slack.md) |\n| Z4 | `Q-tail-derivation-0830` Does the tail's measured law c = 0.7522 ln^2(p'^2) and its surplus derive, and which part of the coefficient is the tile's own (no prime input) against which part needs Hardy-Littlewood? | PARTIAL | PARTIAL, kill clause fired on the coefficient: the ensemble object (the same functional averaged over all p# translates) derives exactly, E_all = R + 5/2 and E_odd = R + 3 with R = Sum g^2/2W, computed exact at x = 7..29 with every conditional the rough-origin candidate needs, and its Mertens constant e^{2gamma}/(8 C2) = 0.6007 is 1/(2 C2) times rho(2) = e^{2gamma}/4 by construction; every route from there to the anchored 0.7574 passes through rho(2)'s limit (the sharp form of Assumption A, algebraically equivalent to HL) and through the zone's gap shape CV^2 -> 1 (an HL statement), so the anchored coefficient is HL in disguise and nothing beyond HL is derived; the tile's own (1 + CV^2)/2 is 0.6306 -> 0.7532 over x = 7..29 and not settled; the record's 0.7522 = HL x 0.9644 x 1.0298 is two non-HL-sized factors cancelling at this height; the rough-origin candidate has an exact ensemble counterpart E_rc/E_cls = 1.0050 -> 1.0523, right sign, wrong size and trend against the anchored 1.0157, not identified; one number derives that the record measured: the class-null offset E_cls - R = 5.600 -> 6.034 exact against the record's 6.05, replacing both the prereg's 0.5 and the note's 7.5; and the corpus's \"R + 1/2\" is the other convention, under the tail's own the constants are 5/2 and 3. | [attack-0830-tail-derivation.md](history/staging/attack-0830-tail-derivation.md) |\n| Z4 | `Q-verify-record-defects-0830` Do the three defect claims of 2026-08-30 (attack-0830-varE-identification.md sec.8 on the doubled X2 column; attack-0830-tail-derivation.md sec.7 on the corpus's \"R + 1/2\"; attack-0830-head-remainder.md sec.1 on the decade-pooled Delta = 0.6214) reproduce on independent code from the definitions, and which of the corrections they list apply? | ANSWERED | Claim 1 CONFIRMED: the corpus delta*X2 column is the positive half of the p \\| h-2 pattern doubled, that pattern's mirror is the p \\| h+2 pattern (W-(-h) = W+(h) exactly, W-(h) != W-(-h) at 9 to 900,679 shifts), the group sum 2X2 reads -0.5603 .. 0.1846 at x = 7..19 against 5.7147 .. 5.3266, and Xmix is overstated 2.305x at x = 19 (6.9x at x = 7); X, X1 untouched. Claim 2 AMENDED: R + 5/2 and R + 3 are exact for the tail convention (all and odd origins, asserted at x = 7..23) and t/(R + 3) = 1.0228, but head-residual-factor.md:70 is the HEAD's forward convention where R + 1/2 and R + 1 are exact, so that correction does not apply; and p'^2 is 1 mod 6, where the tail constant is 5 (t/(R + 5) = 1.0181), with the class null (1.0157) the matched figure, so 1.0228 is one unmatched convention replacing another. Claim 3 CONFIRMED as an artefact, AMENDED on mechanism and on the half-decade reading: 0.6214 pooled against 0.6592, 0.6705, 0.6736 at 2, 4, 8 sub-windows; the between-slope (Simpson) piece the note derives is half of the pooling term (0.0187 of 0.0378 at halves, 0.0261 of 0.0522 at eighths), the other half is the within slope's own fall with height weighted by Var(g); and re-reading the halves from their eighths puts HL BELOW the measurement at 5 of 6 half-decades by 2.5 to 3.4 percent at the top decade, so \"within 1.6 s.e., sign alternating\" is itself a pooled statement. | [verify-0830-record-defects.md](history/staging/verify-0830-record-defects.md) |\n| Z4 | `Q-zone-tail` What is the tail field of the zone (p, p'^2) - its law against ln^2, its worst case, its share of the R0 width, and is it the frozen-sieve last survivor below p'^2 at the rate the head is the frozen sqrt(p)-level first survivor above p? | PARTIAL | The field exists now, 1,225 zones, and nothing in it is derived. The tail is the head's own renewal functional read at height p'^2 rather than p: c_tail = 0.7771 ln^2(p'^2) at [3163,1e4) against the head's 0.6693 ln^2 p on the same zones and the same estimator, both referenced to HL's 0.7574, and the apparent factor of four is the unit ln^2(p'^2) = 4 ln^2 p and nothing else. The band drift does not settle (0.7518, 0.7288, 0.7771; the pre-registered 0.03 step is missed), so this is band composition on two decades and not a measured convergence. The frozen-sieve mirror is 100% at the zone's own level p, but that is the Zone Restriction Lemma restated and not a rate comparable to the head's 96.65%; the non-vacuous mirror is one fold back, where the tail survives at 98.69% with the residual priced exactly (88 extra lpf(n) = p survivors, 18.18% of them biting), and the head's own frozen level sqrt(p) identifies the tail in only 22.86%, refuting this note's own pre-registered [0,5] zones. The renewal surplus t/R = 1.0619 at B4 is NOT RESOLVED, its bootstrap [0.9937, 1.1321] containing 1, and an exhaustive class-matched origin null over 5,962,057 origins gives the same 1.0619; three matched non-square endpoints see no difference at all, sign tests z = 0.14, -0.54, 0.66 with six paired intervals all containing zero, so nothing here is a property of the endpoint being a prime square. R0 is asserted at all 1,225 zones and the shares are 1.82% head, 89.72% Z2, 8.46% tail at B4 with the tail's share FALLING, so filling the tail row moves the decomposition's difficulty nowhere: Z2 was the obstruction before and is more of it after. | [zone-tail-01.md](history/staging/zone-tail-01.md) |\n| Z4 | `Q-zone-tail-02` On the full zonegap-01 range to 1e11, what is the per-zone tail law in both units, is the renewal surplus t/R real at 27,292 zones, and do the head and the tail of a zone have a joint law? | ANSWERED | Resolved DOWN and nothing is derived: zone-tail-01's unresolved t/R = 1.0619 at 782 zones reads 1.0298 on a height-matched comparator at 17,700 zones (bootstrap [1.0156, 1.0443], excluding 1) and 1.0157 against the exactly matched discrete class null (bootstrap [1.0013, 1.0308], clearing 1 by 0.0013), so part of the original figure was the comparator and the pre-registered rule returns MATCH by two parts in ten thousand; c_local = 0.7522 at the top band with bootstrap [0.7410, 0.7630] now CONTAINING HL's 0.7574 and the band drift inside the registered 0.03 per step for the first time but still non-monotone (0.7518, 0.7664, 0.7416, 0.7522), so no convergence is measured; global worst tail/ln^2(p'^2) = 6.840 at p = 15107 with max/mean no longer flat (4.40, 5.36, 8.59, 8.43); Q5 is answered in the negative, head and tail showing no dependence at 27,292 zones (detrended r = -0.0009 against null s.e. 0.0061, permutation p = 0.8920, variance ratio 0.9990, no KS rejection, no chi-square near threshold), so the R0 pieces are a product structure to first and second moments and in the sum's whole law while independence is NOT proven and the simultaneous-extremes regime is untouched; a candidate rough-origin mechanism for the surplus is HEURISTIC and does not fit the head's own excess band by band; the head field's effective sample size is measured at 1,910 distinct a_first among 17,700 top-band zones, so this note's own head interval is priced over draws the field does not hold, a cluster bootstrap over distinct a_first (producer SEC I(a), seed 613009117) measuring the inflation at 4.10 rather than the 3.04 the redundancy alone implies and moving the top-band interval to [0.6848, 0.7692], still inside the registered [0.68, 0.78] (correction originating with redteam-0829-measure-b.md section 1c C15); the scope of that defect is that one interval and not the corpus, since destroyer-census-01.md publishes no bootstrap, head-residual-factor.js bootstraps over gaps, zonegap-01.js has none and zone-tail-01.js bootstraps only tail quantities; ten of ten predictions hit, which is a criticism of the pre-registration; and the R0 shares read 1.40% / 92.80% / 5.81% with the tail's share still falling, so Z2 is more of the obstruction after this pass than before it. | [zone-tail-02-0829.md](history/staging/zone-tail-02-0829.md) |\n| Z5 | `Q-applied-0828-litimports` Were the lit/imports red team's corrections applied to the six HELD notes? | ANSWERED | Applied, 64 substantive edits across the six notes: the Saffari-Vaughan range exponent is 6/11 in both import notes, the fixed-M THEOREM in import-fracparts is struck and the cap-36 MEASURED to PROVEN upgrade withdrawn, row 15 loses its theorem column and row 16 is regraded stale, and the six weakened sentences in the four literature notes now read as the audit requires; four claims are declined with reasons, and nine live documents plus two fenced staging notes are recorded here as owing corrections that this pass did not apply. | [applied-0828-litimports.md](history/staging/applied-0828-litimports.md) |\n| Z5 | `Q-applied-0828-registries` Which lit/imports corrections reached the registries? | ANSWERED | 13 edits across six files, plus one appended changelog section: IMPORT-MAP gains rows 15 (no THEOREM, the column struck), 16 (STALE) and 17 (CLOSURE, two-sided) with its three counts recounted to seventeen rows; SEARCH-CONVENTIONS gains the Kourbatov-b, generalized-Dickman and VC-of-multiples rows with each channel's calibration state; import-map-construction's determinantal cell, natal-cap-36's separate-census sentence, one REFUTED row and one PRIOR-ART block are corrected; GLOSSARY, TODO, QUESTIONS, the five loose hyperuniformity documents and two staging notes are left for their holders and listed here. | [applied-0828-registries.md](history/staging/applied-0828-registries.md) |\n| Z5 | `Q-record-deficit` Does the 6.0 percent record-location deficit survive the corrected null, and what is it? | MIXED (lit-kourbatov-shortfall.md: ANSWERED; record-location-null.md: PARTIAL) | Survives (z = -3.3, five window cuts); Z5's premises on 'location' and on n_eff were wrong; corrected-null formula in §4. | [lit-kourbatov-shortfall.md](history/staging/lit-kourbatov-shortfall.md), [record-location-null.md](history/staging/record-location-null.md) |\n| Z5 | `Q-record-mechanism-0830` Which of three candidate mechanisms (sub-Poisson gap dispersion, the tile's exact gap law, Kourbatov's k = 1 conspiracy heuristic transported to k = 2) carries the residual 0.78 to 0.97 of Kourbatov's b, at what derived size, and what fraction of b is left? | PARTIAL | The residual is carried by the twin-prime gap law AT HEIGHT and by nothing else tried: a new sieve to 1e11 measures that law under-dispersed (CV^2 rising 0.7276 to 0.9295 over seven decades) with a far tail steeper than exponential (log-slope 1.0638 at the top decade), and fed into the record null it carries 95.7 to 99.8 percent of the residual (d b_z = 1.0370 and 1.0806 +- 0.0014 against 1.0833, MEASURED on a model, paired, 2000 replicates); the tile's period-wide law (d b_z = 8.10) and Kourbatov-Wolf's k = 1 bin-count heuristic (4.20) overshoot by 4 to 7x and are closed as posed; the transport of a bare CV^2 is family-dependent (0.62 against 1.06 at CV^2 = 0.93), so \"computable from Var/E\" is false as stated; Kourbatov-Wolf's second-order scale a_c/abar tracks the measured CV^2 to within 0.04 at every decade with no parameter and is an ansatz, not a derivation; the shape statistic moves a third of the way toward the data (z_D from -1.64 to -0.97); what is left with no derivation is the gap law itself, and the top height band's b (0.937 measured against 1.69 to 1.98 predicted on 17 records) is the open falsifier. | [attack-0830-record-mechanism.md](history/staging/attack-0830-record-mechanism.md) |\n| Z5 | `Q-record-null2` How much of Kourbatov's shortfall coefficient b does an inhomogeneous-intensity record null carry, how much does a 6Z-latticed null carry, and what is the residual? | ANSWERED | The inhomogeneous null carries essentially none of b (d b_z = -0.0012 +- 0.0001, MEASURED); the 6Z lattice's share is LAW-DEPENDENT, +0.1113 +- 0.0003 under the memoryless rounding-up law against -0.0060 +- 0.0003 under rounding to nearest, so the null side is a bracket of 16.1 to 25.1 percent of b and three quarters or more stays a residual with no mechanism; record-location-null.md section 8's argued 1.7e-4 lattice bound is wrong but no single number replaces it; the 15 percent estimator disagreement is pinned as an exact weighting identity; and the shape assumption behind every b reading is measured at 1.6 sigma and stays open. | [measure-record-null2-0829.md](history/staging/measure-record-null2-0829.md) |\n| Z5 | `Q-redteam-0828-litimports` Do the six notes of 2026-08-28 survive an adversarial pass, and which of their load-bearing claims are wrong? | ANSWERED | Every measured number reproduces on independent code; Saffari-Vaughan Theorem 10's range exponent is 6/11 not 1/11, which voids import-fracparts's one banked THEOREM at fixed M and corrects row 15 twice; four further weakenings, one refuted verdict line, no note refuted whole. | [redteam-0828-litimports.md](history/staging/redteam-0828-litimports.md) |\n| Z5 | `Q-redteam-0830-records` Do attack-0829n-parity-dstar.md (CLOSED, HELD) and attack-0830-record-mechanism.md (PARTIAL, HELD) survive an adversarial pass on independent code, does REFUTED row 95 stand as worded, and is TODO Z5's new title supported? | ANSWERED | Both notes' measurements reproduce on independent code and neither headline is refuted: D* is confirmed EXACTLY, by a BigInt rational simplex written here, at 299 of 376 anchors with 0 contradictions (the other 77 exceed the exact solver's size cap), the old range's 172-figure gate is re-derived exactly at all 43 anchors with 0 disagreements, and an independent 30-wheel sieve to 1e11 reproduces note (b)'s seven decade gap counts, CV^2 and far-tail slopes to every printed digit, with the counts also matching the published pi_2(10^k). Four sentences are wrong or over-stated and are replaced here: REFUTED row 95, note (a)'s ledger, its section 6 and TODO Z5 all attribute the null's 0.025 agreement to the WIDTH law when 0.025 is the ln Q slope difference (the width-slope difference on the same 84 matched anchors is 0.031); note (a) section 3's \"the threshold sits in a gap of at least 0.25 at every anchor\" is scoped in its own producer to the true arm and the null arm's exact gap is 0.1212; P6's HIT rests entirely on the decade-2 collapse that section 5 discounts as a sampler artefact when it discounts kill clause (a); and note (b)'s \"d b_z = 1.0806 +- 0.0014\" carries only the Monte Carlo pairing error, while moving the far-tail slope by 0.02 (against a spread of 0.0174 across six fit windows on the same decade and 0.1196 across decades) moves d b_z from 0.9888 to 1.1720, and continuing the law at slope 1 above the last resolved point still gives 0.7752. REFUTED row 95's CLOSED verdict stands on kill clause (b), which is better powered than the note presents it (paired s.e. 0.0401 against a 0.15 band), and the row needs two clauses reworded; TODO Z5's title is supported for \"measured and not derived\", and the resolved part of the law carries 72 per cent of the residual on its own with the extrapolation carrying the rest. | [redteam-0830-records.md](history/staging/redteam-0830-records.md) |\n| D | `Q-doubling-C2` Does the doubling inequality hold on the base-2 chain, and can a bridging certificate prove it? | PARTIAL | Not proven and not refuted: the exact C2 table has sup 5.2727 at s = 16, eleven proven finite-level bounds C2 <= K*+1 tight to a factor <= 2.91, two closures refuted outright, and the alarm is that K* drifts up (slope 0.6881 +/- 0.1328) while C2 does not (0.2018 +/- 0.1412). | [attack-doubling-01.md](history/staging/attack-doubling-01.md) |\n| D | `Q-doubling-bridge-0829n` Can a bridging certificate carry Ghat(2s) <= 8 Ghat(s) from level s to level 2s uniformly in s, all s, on the base-2 chain? | PARTIAL | Not by any proven mechanism: the K*-product bridge is CLOSED at every C2 by the cited run floor K* >= pi(2s)-pi(s) (K* = 17 at s = 16 by exact walk, VERIFIED, so the certificate reads 18 against 8 on the chain itself; it exits the whole legal band at s = 128, PROVEN); the sharper maxsum bridge Ghat(2s) <= maxsum_{K*+1}(T_s) is PROVEN and holds under 8 at all fourteen enumerable steps (VERIFIED), but its all-s form needs an upper bound on K* against the entering primes' two-class covering that nothing proven supplies; the doubling inequality itself is untouched. | [attack-0829n-doubling-bridge.md](history/staging/attack-0829n-doubling-bridge.md) |\n| D | `Q-doubling-killrun-0830` Can the exact per-fold L bounds of the pi(2s) - pi(s) folds inside one doubling step compose to an upper bound on the weighted kill-run below the allowance 8 Ghat(s)/gbar(s), for all s, without passing through K*? | PARTIAL | No, and the composition is closed as a route, not merely unproven: the folds compose as a PRODUCT, K*+1 <= prod_j (1+L_j) (PROVEN here, exact at fourteen steps, 540 against 18 at s = 16), the composed certificate never sits below yesterday's maxsum certificate (PROVEN by monotonicity), it exceeds the allowance at 8 of 14 enumerable steps starting at s = 9 (VERIFIED), and since every fold kills a slot the composed index is at least 2^N, whose floor 2^N gbar(s) alone exceeds 8 Ghat(s) at the chain rungs 16, 32, 64 and beats the cited polynomial ceiling on Ghat for all large s (PROVEN given the ceiling); the sum form is false at seven steps and the max form is a floor; the weighted run itself, computed exactly, sits at 0.83 of its allowance at s = 16 and (M8) is exactly where yesterday left it, OPEN. | [attack-0830-doubling-killrun.md](history/staging/attack-0830-doubling-killrun.md) |\n| D | `Q-kstar-prereg` What is K* at the three next doubling steps, predicted before any period walk? | OPEN | Pre-registration only, committed alone: the predictions, the scoring rule and the growth-type verdict thresholds are fixed in advance, with the inclusion-exclusion engine validated against an independent scan engine on all eleven known steps first. | [attack-kstar-01-prereg.md](history/staging/attack-kstar-01-prereg.md) |\n| D | `Q-recon-0830-rec-killrun` Does the literature hold, in its own conventions, a theorem whose hypotheses REC(s, u0) (the sup-versus-rms recovery of the level-D signed remainder over all positions, attack-0829n-rml-proof.md section 3) or the weighted kill-run K*(s) (the longest run of level-s slots the primes in (s, 2s] can kill, attack-0829n-doubling-bridge.md sections 0 and 3) satisfies as stated, or a Maier-type theorem that makes REC false for two-class sifted sets at some u0 below beta_2? | ANSWERED | No theorem applies to either object as stated, on four calibrated channels searched in the owning conventions; every neighbour is graded NEAREST with its unmet hypothesis named, and neither object is on any refuted row. On the Maier question the answer is negative for REC as stated (the Maier family bounds the COUNT, i.e. the full-level remainder, never a level-D truncation) but carries one calibration: at one class and full level the REC-shaped inequality is FALSE asymptotically, by Buchstab's origin ratio against the Montgomery-Vaughan full-period variance ceiling, and the embedded arithmetic puts the level where that falsity first shows at log10 z between 16 and 141 depending on u0 and epsilon, so a finite-z margin of the kind the corpus measures cannot see a failure of this type; at two classes the same full-level statement is HL-conditional at u0 = 2 and unlocated in print at any u0. No exponent moved. | [recon-0830-rec-killrun.md](history/staging/recon-0830-rec-killrun.md) |\n| D | `Q-redteam-0830-doubling` Do attack-0829n-doubling-bridge.md and attack-0830-doubling-killrun.md survive an adversarial re-derivation on an engine sharing nothing with their producers, and do REFUTED rows 94 and 98 stand as worded? | ANSWERED | The mathematics survives: the maxsum certificate Ghat(2s) <= maxsum_{K*+1}(T_s) and the product composition K*+1 <= prod(1+L_j) are each re-derived here and hold, the second at 21,641,346 nesting links over 6,012,804 killed runs with zero failures including the zero-kill fold, and every quoted figure reproduces digit for digit on a fresh engine (K*(16) = 17; G2(31#) = 348 @ 8813641451 x4 from both base tiles; sup msc 6.6364; the diagonal cells; the 2^N ratios 1.226/1.481/41.348). Row 94 STANDS and is if anything under-claimed, but its attribution is wrong in one direction: Lemma 1 at s = 128 alone clears the whole band, so the s = 16 walk corroborates rather than carries it. One statement is falsified: the bridge note's NOT-REACHED line puts 19#->37# out of reach, but column-major it returns here, reproducing the ladder row x = 37 and adding a FIFTEENTH step at s = 19, 20 (K* = 13, N = 4, C2 3.5200, certificate 3.8000), so the enumerable range ends at s = 20, not s = 18. Row 98 STANDS on its mathematics and is WEAKENED on one clause: \"the truth sits under it everywhere (sup w/a = 0.8295)\" attaches the CERTIFICATE's ratio to the word truth; the truth's sup is 0.6591. | [redteam-0830-doubling.md](history/staging/redteam-0830-doubling.md) |\n| D | `Q-redteam-0830-imports` Do the six lighter notes of 2026-08-30 (varE-identification, smooth-aps, rec-killrun, skeleton-door, at43-bigint, coherence) survive an adversarial pass on independent code and at the source pages, and which of their load-bearing claims are wrong? | ANSWERED | No note is refuted whole and no exponent moves. Every decisive measurement reproduces on code sharing nothing with the producers: the varE mixed remainder and its 0.844 / 0.826 share below 2L and 94.4 / 98.7 percent unbalanced share at x = 13, 17; the Buchstab deviations and all three crossing levels log10 z* = 33.53, 68.80, 33.62; the skeleton door's 9.2 / -0.9 percent, G30_agg 0.1113 / 0.1011 and the 0.0102 cap at @13, @17, by a path that also verifies cap-36 Theorem A; the @43 parity exactness on all 26,157,448 scour classes with 0 above 2^54 and a constructed @47 counterexample; and the five recon-0828 verdict counts summing to 75, which REFUTES rec-killrun's brief-error clause and confirms the orchestrator's rider. Seven claims WEAKEN. The load-bearing one is smooth-aps's \"the weight is unmet in print\": Harper's J. London Math. Soc. 112 (2025) e70293 proves a Barban-Davenport-Halberstam asymptotic for an ARBITRARY sequence on exactly that dyadic all-classes object, and his own footnote says the missing max over x' can be incorporated, so the NEAREST citation changes even though the step still does not close (the 2025 range is sqrt(2x) < Q <= x and its Theorem 2 route bars sieved sets by name). Also: Henriot Corollary 2 applies, but to the ERRATUM's statement and under two hypotheses the note leaves unstated (eps < alpha/600, not \"every eps\"; F at prime powers); varE section 4's display (*) carries an unfilled placeholder and is vacuous by a factor L/d; the skeleton door's \"would tighten\" is wrong in sign at @17, where the closable block is negative and removing it RAISES G30_agg to 0.1020; Halasz-Montgomery is not applicable for the reason given (it is an abstract inner-product inequality); FGKT's heuristic quote is on p. 4, not p. 3; and the \"~13x had no source\" is a corpus miss, the retired G2-walk figure at CHANGELOG.md:5716. Not TPC-strength; nothing here opens a route. | [redteam-0830-imports.md](history/staging/redteam-0830-imports.md) |\n| 0 | `Q-applied-0828-closures` Were the closures red team's corrections applied to the five HELD notes? | ANSWERED | Applied. 47 prose corrections across the five notes; c2prime-refit-22 drops from CLOSED to PARTIAL and three ledger verdicts are rewritten; four producer READINGS blocks and eleven live-document lines carry the same defects and are listed here uncorrected, behind the fence. | [applied-0828-closures.md](history/staging/applied-0828-closures.md) |\n| 0 | `Q-barrier-kappa2` Does Granville's exceptional-character construction, which makes the Jurkat-Richert f and F attained by interval sieving problems at kappa = 1, transfer to kappa = 2 and supply the missing barrier on wall-note Face 4, so that no class-blind argument can beat beta_2 = 4.26645 on intervals? | CLOSED | No. Verdict (b): the transfer runs, fixes the axiom obstruction that sift-limit-attack.md section 2 blamed for the mixed-sign sketch's death, and then hits a hard ceiling at u = 2, because the character's only gift is the one-point identity lambda = chi on z-rough integers and a sign pattern with a +1 coordinate is empty exactly when z^2 >= x; above u = 2 the surviving question is the shifted intersection of two positive-density lambda-classes, which is Chowla-strength, so the section 2 verdict stands with a corrected reason; the band (2, 4.26645] is untouched, and the hypothesis itself implies the target by Heath-Brown 1983. | [attack-barrier-kappa2.md](history/staging/attack-barrier-kappa2.md) |\n| 0 | `Q-derive-0904-L7-transfer` Is the legal open-set target L7, G2(x#) << g(x#) (ln x)^A for a fixed A (which would give exponent 2 + o(1) from Iwaniec's one-class bound, below beta2 and above the TPC line), reachable by a TRANSFER from the one-class Jacobsthal bound, and if not, what does each transfer mechanism reduce to? | PARTIAL | No transfer mechanism in the corpus reaches L7, DERIVED: the union-bound transfer is vacuous already in the period average from x = 11 (the exact CRT budget sum of 1/(p - 1) over 3 <= p <= x passes 1 at x = 11), and the sieve-on-holes transfer is the Bruedern-Fouvry vector sieve, whose coupled unconditional form lands at 2(1 + sqrt e) = 5.2974 above beta2 and whose decoupled form needs the sup-over-positions remainder law that item 0 ruled a truth gap; L7 stays a legal OPEN statement with no mechanism, and the conjectural A = 1 is truth-side evidence only. | [derive-0904-L7-transfer.md](history/staging/derive-0904-L7-transfer.md) |\n| 0 | `Q-g2-provenance-x37` Does the headline G2 exponent depend on the eight terms this repository has never reproduced, and does any instrument without G2(37#) inside it flag x = 37? | ANSWERED | MEASURED, both largely negative: the custody-only refit lands at 1.533 against the headline 1.498 and no provenance-block drop moves the corrected central by more than 0.035 (Ghat(64) alone moves it 0.002), while no G2-free instrument flags 37 in the high direction, the one \\|z\\| past 2 being h(37#) at -2.20 in the low direction and range-dependent; the blind extreme-value forecast of G2(37#) does overshoot at z = +6.58, unexplained. | [measure-g2-provenance-0829.md](history/staging/measure-g2-provenance-0829.md) |\n| 0 | `Q-gap-spectrum` What is the full gap-length distribution of twin slots over one period of the tile, what does the excess functional \\|B_N\\| = sum_{g>N}(g-N) read on an x^s grid, and where does the empirical minimum window count leave zero? | ANSWERED | The distribution is exact at nine levels to x = 31 and the pre-registered renewal null FAILS in the light direction at every one: the tail is uniformly steeper than exponential (body slope -1.3302 against -1), no power law survives (the power-law slope moves by a factor 17), the maximum is 2.087 to 2.241 BELOW the renewal maximum, and \\|B_N\\| sits at 0.0245 to 0.7436 of the null and is exactly zero at s = 1.75 where the null predicts 6.5262e+5 empty windows; so no heavy-tail mechanism for a super-linear G2 exponent exists at any computable level, the identity G2 <= N + \\|B_N\\| is proven but degenerates to G2 <= N on this grid, and the sifting onset is flat at 1.63 to 1.70 against a sieve bound f2 that is zero everywhere the grid reaches. | [gap-spectrum-01.md](history/staging/gap-spectrum-01.md) |\n| 0 | `Q-lit-siebert` Is Siebert's 1983 result stated for general sieve dimension kappa, or for kappa = 1 only, and does it therefore already contain the kappa = 2 extremal example that SEARCH-CONVENTIONS section 3 records as not found? | ANSWERED | kappa = 1 ONLY, on three independent statements of the theorem (the chapter's own publisher abstract, Granville 2020 section 1, Friedlander-Iwaniec 2022 page 20), all naming the LINEAR sieve and the Jurkat-Richert f and F; the chapter's full text was NOT reached, so this is testimony not primary reading, and the SEARCH-CONVENTIONS section 3 \"no published kappa = 2 extremal example\" row needs no correction. | [lit-siebert-1983.md](history/staging/lit-siebert-1983.md) |\n| 0 | `Q-maximal-law-tpc` Is the SIEVE Gaussian maximal law TPC-implying? | ANSWERED | YES in the only form that would deliver anything: stated for the operative remainder R_H, need/z^2 runs 0.49 to 0.61, flat and below the zone budget at every z from 13 to 43, which is the Gap Reformulation, hence the Zone Postulate, hence TPC, so no soft argument can prove it; the published above-2 theta column is not the maximal law's verdict but the law plus two lossy steps. | [phase1-T4-maximal-law.md](history/staging/phase1-T4-maximal-law.md) |\n| 0 | `Q-measure-g2-generic` Where does G2(x#) sit in the distribution of the same maximum gap taken over other two-class configurations with the same census, and is the constant-shift family generic inside the free-pair family? | ANSWERED | G2 is GENERIC in this coordinate on the registered rule, MEASURED: the midrank percentile of G2 in the census-matched two-class ensemble reads 20.00, 18.06, 26.35, 17.14 exhaustively at x = 11..19 and 13.70 sampled at x = 23, all inside the registered [5, 95] band so the falsifier does not fire, while sitting about one ensemble sd below the mean at all five levels; and the free-pair and constant-shift ensembles are PROVEN identical up to translation, so killer 1 gains no target in the class positions at these levels. | [measure-g2-generic-0829.md](history/staging/measure-g2-generic-0829.md) |\n| 0 | `Q-omega-floor-blind-0830` Does the local log-slope of the pointwise floor Omega(z) of attack-0830-rec-cheapest.md (exact to z = 73, certified 2+4-chain family to z = 5e5), extended blind to new levels under a sealed forecast, behave as the DERIVED growth law Omega >> z^{16s/9}/ln^8 z predicts at s = 2.698721, and at what z would the test be decisive? | ANSWERED | MEASURED, blind, on the derived law only: 18 of 22 sealed rows HIT; the certified 2+4-chain family at p* = (47, 43) runs 5e5 -> 1e9 (A1A2 = 8.863e17 -> 9.351e32, exact integers) with its step slope falling 4.85 -> 4.38 through every sealed band onto the law's log-corrected slope, the decisive decade 1e8 -> 1e9 reading 4.410 against 16s/9 - 8/ln z = 4.389 (HIT, kill rule not triggered at 4.452 +/- 0.011); the 4 MISSes are the ratio to z^{16s/9}/ln^8 z sitting at 2.00-2.14 above the sealed cap of 2 at z >= 1e8, the pre-declared non-adverse direction; the exact floor is extended 73 -> 113 (251 -> 1980, equal to the hill-climb floors at 89 and 101), where the law predicts nothing; no exponent moves and the law stays DERIVED and HELD. | [blind-0830-omega-floor.md](history/staging/blind-0830-omega-floor.md) |\n| 0 | `Q-rec-cheapest-0830` At the cheapest legal point of REC(s, u0), s -> 1+sqrt(e) and u0 -> beta_2, where the saving asked over the triangle inequality is only z^{1.031}, does any of the five dead routes close, does a mixed mean-square-plus-cap strategy close, and is the failure a proof gap or a truth gap? | PARTIAL | No route closes and the failing step is yesterday's statement, so PARTIAL on the routes; but the step is located as a TRUTH gap, not a proof gap, asymptotically: the certificate's own value at a CRT-planted doubly-smooth window is cc(r) = -(A1A2 + A1B2 + B1A2) with every term a count of exit chains of the Rosser supports (PROVEN, checked exhaustively at z <= 73), the window bound T <= cc(r) + H - 1 is exact, and the exit-chain count is DERIVED (held) to grow like z^{16s/9 - o(1)} at four primes and z^{2s - o(1)} in the limit, so REC(s, u0) is false for every u0 < 16s/9 = 4.7088 at s -> 1+sqrt(e), the whole legal band (2, beta_2] included, and RML(alpha) with it; at every computable z it is a proof gap (Omega(z) exact to z = 73 reads 251 against z^{u0} = 1.05e6, and a certified lower bound 8.86e17 at z = 5e5 sits a factor 1.2e6 under z^{u0}); no exponent moved. | [attack-0830-rec-cheapest.md](history/staging/attack-0830-rec-cheapest.md) |\n| 0 | `Q-recon-0828-covering` Does the interval-covering, automated-proof or extremal-combinatorics literature hold a method that bounds the length of an interval covered by two residue classes per prime, or that turns the exact G₂ ladder into a statement at all x? | ANSWERED | No route. Fifteen angles examined, fourteen dead on arrival or already closed here; the best interval-covering bound in print is Crittenden-Vanden Eynden's own Lemma 2, which reads 1.02e12 against G₂(79#) = 1710 and is 8181x weaker than the corpus's own 4.2665 sieve bound at the same level; the one survivor, transplanting Costello-Watts's recurrent certificate to two classes, is a computation that certifies finitely many levels and whose analytic half is Erdős #970 at one class. | [recon-0828-covering.md](history/staging/recon-0828-covering.md) |\n| 0 | `Q-recon-0828-farfields` Do any of the fields still absent from IMPORT-MAP.md hold a theorem whose hypothesis the tile satisfies exactly and whose conclusion is a lower bound on a periodic set's count in every window, a maximal-gap bound for a product-structured periodic set, or a positivity statement at a distinguished point? | ANSWERED | No route from any of the sixteen candidates priced; two reach STRONG-ANALOGY or better and CLEAN and both bank only a closure or a wall address, the Fourier one by a one-line non-vanishing proof that empties the Donoho-Stark/Tao/Meshulam family on this object, and Janson's lower tail by saturating at exp(-1/c) >= 0.0805 against the 1/W a position union bound needs, short by 3.6e2 at x = 11 and 2.8e13 at x = 41. | [recon-0828-farfields.md](history/staging/recon-0828-farfields.md) |\n| 0 | `Q-recon-0828-rough` Does the 2000-2026 literature on y-rough numbers, sifted sets and short intervals hold any theorem that lowers Z2 or G2, meaning any control on the largest gap or the minimum count of a sifted set in EVERY short interval, or any structure of rough numbers a gap bound could use? | ANSWERED | No. Ten angles examined, ten killed for this purpose; the strongest every-interval statement in print for one-class rough numbers is Iwaniec's S(x,y,z) >= (4y/log^2 y)(log(y/z^2) - O(1)) for y >> z^2, uniform in x, giving J(P(z)) << z^2, and it is the linear sieve run exactly at its sifting limit beta_1 = 2 with no slack outside it; no two-class analogue exists in print at any exponent, the corpus's own x^4.2665 is beta_2 by the same mechanism, and every remaining branch of the field (Matomaki-Radziwill, Gorodetsky, Friedlander-Iwaniec Cor 6.28, Gafni-Tao) carries the almost-all quantifier, which a second-moment statement can never convert into a bound on one empty window. | [recon-0828-rough.md](history/staging/recon-0828-rough.md) |\n| 0 | `Q-recon-0828-sieve` Is there anything in the 2008-2026 sieve literature, or in the parity-breaking literature, that lowers beta_2 = 4.26645 or supplies a consumer for a level theta > 1 on the two-class comb? | ANSWERED | No route. Twenty-one angles examined, nineteen dead on arrival, nothing in print beats 4.26645 at kappa = 2 at any level on four calibrated channels; the two survivors are an owning-convention row and one unresolved hypothesis in a five-page 1990 announcement whose stated requirement beta < 3 excludes beta_2 = 4.26645. | [recon-0828-sieve.md](history/staging/recon-0828-sieve.md) |\n| 0 | `Q-recon-0829-escapes` Each of the three killers is an obstruction shape that occurs elsewhere in mathematics; where has each shape been beaten in print, by what mechanism, under what hypothesis, and does that hypothesis hold for the tile and zone objects? | ANSWERED | Of nineteen escapes read, fourteen fail a hypothesis check (twelve on a fact already in this corpus), four fail some other way and one (the decoupled vector sieve) stays OPEN as the programme's existing road, so no route opens; Granville-Soundararajan's uncertainty principle is NOT killer 3 in print, is adverse-direction, and is already cited here twice. | [recon-0829-escapes.md](history/staging/recon-0829-escapes.md) |\n| 0 | `Q-recon-0829-farfields2` Do any of eight fields still absent from IMPORT-MAP.md and from the three 2026-08-28 recon notes hold a theorem whose hypothesis the tile or the zone satisfies exactly and whose conclusion is a lower bound on a periodic set's count in every window, a maximal-gap bound for a product-structured periodic set, or a positivity statement at a distinguished point? | ANSWERED | No route from any of the eight; the pass banks one PUBLISHED-ANCHOR (Holt and Rudd's transfer matrix has its full spectrum in print, second eigenvalue a2 = prod_{17<=q<=p}(q-3)/(q-2), printed 0.10206751799779 at p = 999999999989, decaying only like 2.82/ln p, and this corpus's prior-art record names the eigenstructure and the binomial eigenvectors but carries the per-prime eigenvalue (p-j-1)/(p-2) at lit-pdf-holt-rudd.md line 259 but no product-over-primes closed form, no numerical value and no rate), one CLOSURE with mechanism (Beurling-Selberg is the proof device behind the large sieve, whose constant N-1+1/delta is an l2-to-l2 statement and whose error 1/delta exceeds the window at the operative theta_total > 2, so the whole extremal-function family reproduces a row already rejected), and three wall addresses; two angles are TPC-strength at the form that would pay and are labelled (ii). | [recon-0829-farfields2.md](history/staging/recon-0829-farfields2.md) |\n| 0 | `Q-recon-0830-rec-killrun` Does the literature hold, in its own conventions, a theorem whose hypotheses REC(s, u0) (the sup-versus-rms recovery of the level-D signed remainder over all positions, attack-0829n-rml-proof.md section 3) or the weighted kill-run K*(s) (the longest run of level-s slots the primes in (s, 2s] can kill, attack-0829n-doubling-bridge.md sections 0 and 3) satisfies as stated, or a Maier-type theorem that makes REC false for two-class sifted sets at some u0 below beta_2? | ANSWERED | No theorem applies to either object as stated, on four calibrated channels searched in the owning conventions; every neighbour is graded NEAREST with its unmet hypothesis named, and neither object is on any refuted row. On the Maier question the answer is negative for REC as stated (the Maier family bounds the COUNT, i.e. the full-level remainder, never a level-D truncation) but carries one calibration: at one class and full level the REC-shaped inequality is FALSE asymptotically, by Buchstab's origin ratio against the Montgomery-Vaughan full-period variance ceiling, and the embedded arithmetic puts the level where that falsity first shows at log10 z between 16 and 141 depending on u0 and epsilon, so a finite-z margin of the kind the corpus measures cannot see a failure of this type; at two classes the same full-level statement is HL-conditional at u0 = 2 and unlocated in print at any u0. No exponent moved. | [recon-0830-rec-killrun.md](history/staging/recon-0830-rec-killrun.md) |\n| 0 | `Q-recon-0904-sifting-limit-floor` What is actually known, in the owning convention and at the page, about LOWER bounds on sifting limits at dimension kappa > 1, and is the class-blind cap at beta_2 = 4.26645 a barrier or a method artefact? | PARTIAL | Both. The corpus's [ABSENT] is wrong for lower bounds and right for extremal examples: Selberg's Lectures section 17 and Brady's 2017 Corollaries 1 and 3 ARE lower bounds on beta(kappa) at kappa > 1, invisible because Selberg's convention is the reciprocal a_k = 1/beta_kappa so his lower bounds are titled \"upper bounds for sifting limits\", while no kappa > 1 extremal example exists and Halberstam's own 2003 review says so; the best floor on beta(2) resting on a citation is 1.8394 from Brady's reduction evaluated at kappa = 2, unconditional and below the band, and beta(2) >= 2 follows from monotonicity by a padding argument DERIVED here and probably folklore; the cap at 4.26645 is a METHOD ARTEFACT at rung MEASURED, since the level-D LP is exactly the axiom-only information class and its calibrated floor of 3.3152 sits about 18 per cent below the DHR value and passed, 0 of 15, its first external test against the published beta_3 and beta_4; the band (2, 4.26645] is untouched by anything proven. | [recon-0904-sifting-limit-floor.md](history/staging/recon-0904-sifting-limit-floor.md) |\n| 0 | `Q-redteam-0828-closures` Do the five 2026-08-28 closure notes stand, and do they stand for the reasons they give? | ANSWERED | Three stand on corrected sentences (1d, Z5b, 0); one stands in half (10, the 1.074 counter-claim does not); one does not stand for its stated reason (1c, closed on a narrower window than the prior work's significant one, and its CLOSED has overwritten a PARTIAL in the generated index). | [redteam-0828-closures.md](history/staging/redteam-0828-closures.md) |\n| 0 | `Q-redteam-0830-floor-growth` Does the DERIVED growth half of attack-0830-rec-cheapest.md sec.4.2 (Omega >> z^{16s/9}/ln^8 z from four-prime Rosser exit chains) survive an adversarial re-derivation on independent code, does it really make REC(s, u0) false for every u0 < 16s/9, and does the blind slope test of blind-0830-omega-floor.md test the LAW or the CONSTRUCTION? | ANSWERED | No break found in six attacks, so the growth half SURVIVES this pass and stays DERIVED (one failed break is not a proof, and the whole thing still rests on the exact half E1-E3 that this pass did not grade); but four claims are corrected - the threshold 16s/9 needs the unstated hypothesis s <= 3 (above it the four-prime LP maximum is 2(2s+2)/3, then 8, both below 16s/9, though the corollary that every legal u0 < beta_2 fails survives at every admissible s because 2 x LP-max > beta_2 for all s > 9beta_2/16 = 2.3999), the 2s - o(1) limit needs 3^k <= ln D / ln 8 so only k = 2 is available at every z below 1.1e9, A_i is >= not ~ the dyadic count, and the certified ratio does NOT rise monotonically (0.0118, 0.0128, 0.0101, 0.0057, 0.0051, 0.0093, 0.0336, 0.0801, 0.1658, 0.3545 on my own recount, which reproduces all ten S4b rows digit for digit); the blind test is ruled a test of the CONSTRUCTION's count and not of the law, since it cannot see E1-E3 or the CRT step; on the ruling Chris asked for, the floor does make the RML route a truth gap asymptotically and conditionally on E1-E4, and changes nothing computable, the literal construction first passing H = z^{u0} at z ~ 5.6e31 on the constant 2.93e-4 measured here; no exponent moved. | [redteam-0830-floor-growth.md](history/staging/redteam-0830-floor-growth.md) |\n| 0 | `Q-redteam-0830-floor-sign` Does the EXACT half of attack-0830-rec-cheapest.md sec.4 (E1-E4, the sign facts, the window bound T <= cc(r) + H - 1, the exact Omega, and sec.(d)/(e)'s calibration sentences) survive an adversarial re-derivation on independent code, and does it by itself constrain REC at any u0 without the growth law? | ANSWERED | The mathematics STANDS as PROVEN and is stronger in two places than the note claims, but it constrains REC at NO u0 by itself: E3's identity is pure algebra and holds at every position (0 mismatches in 20.5M positions, both s), E2 holds at every divisor n > 1 under three disjoint routes at every s >= 1 and FAILS at s = 0.8, locating a hypothesis D >= z the note omits, E4 has 0 violations at every x over four complete periods with the wrap, and Omega is reproduced by an independent enumeration at all 26 tabled values and EXTENDED to z = 101 (285, 441, 684, 990, 1155 at z = 79, 83, 89, 97, 101), confirming blind-0830's exact 684 and 1155; three sentences are wrong -- E1's \"every pair with gcd \\| 2 is realised\" (exactly half are; the missing half is all those with 2 dividing one side), the ledger's and sec.0's \"251 against z^{u0} = 1.05e6\" (1.05e6 is HM, z^{u0} = 7.19e7, so the gap is understated 68.3x), and sec.4.4's reason for lam+(P(z)) = 0 (0.24 exceeds 1/8; the figure that closes the slab is 0.1199 at z = 89, under 1/8 by 4 percent) -- and sup\\|rho~\\| >= (Omega - M)/2 is loose: sup\\|R_1\\| = Omega + M exactly. Every exactly known log_z Omega is under 2 while REC needs it above u0 > 2, so the exact half touches REC at no u0 and every adverse consequence in the note still rests entirely on the growth half. No exponent moves. | [redteam-0830-floor-sign.md](history/staging/redteam-0830-floor-sign.md) |\n| 0 | `Q-redteam-0830-imports` Do the six lighter notes of 2026-08-30 (varE-identification, smooth-aps, rec-killrun, skeleton-door, at43-bigint, coherence) survive an adversarial pass on independent code and at the source pages, and which of their load-bearing claims are wrong? | ANSWERED | No note is refuted whole and no exponent moves. Every decisive measurement reproduces on code sharing nothing with the producers: the varE mixed remainder and its 0.844 / 0.826 share below 2L and 94.4 / 98.7 percent unbalanced share at x = 13, 17; the Buchstab deviations and all three crossing levels log10 z* = 33.53, 68.80, 33.62; the skeleton door's 9.2 / -0.9 percent, G30_agg 0.1113 / 0.1011 and the 0.0102 cap at @13, @17, by a path that also verifies cap-36 Theorem A; the @43 parity exactness on all 26,157,448 scour classes with 0 above 2^54 and a constructed @47 counterexample; and the five recon-0828 verdict counts summing to 75, which REFUTES rec-killrun's brief-error clause and confirms the orchestrator's rider. Seven claims WEAKEN. The load-bearing one is smooth-aps's \"the weight is unmet in print\": Harper's J. London Math. Soc. 112 (2025) e70293 proves a Barban-Davenport-Halberstam asymptotic for an ARBITRARY sequence on exactly that dyadic all-classes object, and his own footnote says the missing max over x' can be incorporated, so the NEAREST citation changes even though the step still does not close (the 2025 range is sqrt(2x) < Q <= x and its Theorem 2 route bars sieved sets by name). Also: Henriot Corollary 2 applies, but to the ERRATUM's statement and under two hypotheses the note leaves unstated (eps < alpha/600, not \"every eps\"; F at prime powers); varE section 4's display (*) carries an unfilled placeholder and is vacuous by a factor L/d; the skeleton door's \"would tighten\" is wrong in sign at @17, where the closable block is negative and removing it RAISES G30_agg to 0.1020; Halasz-Montgomery is not applicable for the reason given (it is an abstract inner-product inequality); FGKT's heuristic quote is on p. 4, not p. 3; and the \"~13x had no source\" is a corpus miss, the retired G2-walk figure at CHANGELOG.md:5716. Not TPC-strength; nothing here opens a route. | [redteam-0830-imports.md](history/staging/redteam-0830-imports.md) |\n| 0 | `Q-redteam-0830-rml` Do the three HELD notes of 2026-08-29 night and 2026-08-30 on the RML chain survive an adversarial pass — the six-arrow chain and the term-count cap lemma of attack-0829n-rml-proof.md, the same-object ruling and the applied replacement sentences of verify-0830-usup-convention.md, and the reflection identity, custody, pilot bounds and sealed forecast of measure-0830-rho-sup-z41.md? | ANSWERED | The three load-bearing PROVEN claims survive independent re-derivation — the CAP lemma (every step, sine sum at worst ratio 0.585568, the constant 6, 0 per-modulus violations at z = 13, 17), the reflection identity rho~(W-3-y) = -rho~(y) (proved here and its two hypotheses verified at z = 13..37, residual <= 1.7e-10 at every position of z = 13, 17, 19), and the same-object ruling (an independent lattice reproduces the repository's term multiset exactly at z = 13..37 and its {Vabs > 0} set equals LSE's records at all ten levels); five of the seven cited exact sups and three of the five walked sup\\|R_H3\\| reproduce to <= 5.0e-7, and every figure of the sealed z = 41 forecast reproduces from the seven cited sups alone. Fourteen claims do not survive, none of them reopening a route. The load-bearing one is RML sec.2's ms_m/ask column: it measures the PROVEN mean-square BOUND sqrt(B H), not the ^{1/2} that REC names, the two differing by 36.53x to 200.45x on the tabled rows because saturates in H while B H does not, so sec.5's ordering (ask below the Gaussian factor at u0 = 3.0 and 3.5) is an artifact and reverses at every row against the true mean square. \"gauss -> 0\" (sec.5) and \"sup/rms <= sqrt(2 lnW) O(1) = z^{o(1)}\" (sec.3) are the same error twice, against the note's own sec.0: sqrt(2 ln W) = z^{1/2+o(1)}, measured gauss 0.7905 at z = 19 falling to 0.5645 at z = 1e7, limit 1/2. \"the recovery exponent is log_z(2 sup\\|rho~\\|/rms): about 0.99-1.29\" quotes the truth column, which is log_z(2 sup\\|rho~\\|) alone; the quantity named measures 0.5628 to 0.6712. \"REC would be moot rather than false\" should read false. \"u_cap -> 2s + o(1)\" and \"n ~ z^{2s}/polylog\" are asserted where only \"<=\" is proven and the data cannot decide: u_cap is 4.2652 at z = 47, crosses beta_2 only at z = 53, reaches 4.4891 at z = 79, and passes u_triv from z = 53. \"seven points spanning less than one octave\" is 1.509 octaves, still uncorrected in three places including the ledger verdict. verify-0830's applied \"polynomially many from z = 223 on\" is true at every z >= 13, and its z = 331 crossover silently sets A = 1 (A fitted to LSE's own Ssat moves it to 337). No exponent moves, no route opens or closes, and no ledger verdict changes. | [redteam-0830-rml.md](history/staging/redteam-0830-rml.md) |\n| 0 | `Q-redteam-0904-floor-growth-2` Does the growth half of the adverse pointwise floor (Omega >> z^{16s/9}/ln^8 z from four-prime Rosser exit chains) survive a SECOND adversarial pass, and does a floor on the certificate at a CRT-planted window make REC(s,u0) false (a truth gap) or only unprovable by that certificate (a proof gap)? | ANSWERED | SURVIVES, at the same rung and no higher: four new attacks and two unasked ones found no break, so the growth half stays DERIVED and red-teamed twice, which is two failed breaks on a derivation and not a proof; the proof-gap reading is REFUTED because REC is a statement about R_H, the remainder of the very certificate the floor bounds below, and the chain has no step where a certificate value must become a true remainder, so it is a truth gap for REC as the corpus states it; five corrections are owed to the record -- the REFUTED row is UNSCOPED and should read \"on the Rosser vector-sieve lattice at equal levels\" since only the Rosser instance is derived, the \"falsity beyond z ~ 5.6e31\" is an s = 2.698721 figure quoted without its s and the fitted crossing at the corpus's own measured family s = 3.0 is z ~ 10^15.3, the analytic p* window (4D^{1/27}, D^{1/9}/4) is neither necessary nor sufficient at reachable levels so the exact onset is 4.42e5 at s = 3.0 rather than the quoted 1.055e6, the four-prime construction has NO exit prime below z once s >= 7 so the first pass's \"8 beyond s = 5\" holds only on [5,7), and the first pass's escape-(a) density 1/W understates the planted class by the factor W/(n1n2) though the escape still fails by an exponential; the LP maximum 8s/9 is confirmed as an exact vertex and the corollary crossing reproduces 9beta2/16 to 4.4e-16, the pre-registered slope band held at a new s (top step 4.9025 at s = 3.0 against a model 4.8990), and no exponent moved. | [redteam-0904-floor-growth-2.md](history/staging/redteam-0904-floor-growth-2.md) |\n| 0 | `Q-redteam-0904-item0` Do the two HELD notes of 2026-09-04 on item 0, the second adversarial pass on the adverse floor's growth half and the L7 one-class-to-two-class transfer, survive an independent adversarial pass, and do the live-layer changes they propose survive verbatim? | ANSWERED | Both notes SURVIVE at their own rungs and neither moves an exponent, but SIX of their proposed live-layer changes need rewording and two of their supporting derivations are wrong; every load-bearing number reproduces on independent code (18 of 18 A1A2 cells digit for digit, the LP maximum 8s/9 at 0 deviations of 16 with residual 4.441e-16, the crossing 2.399878284834 against 9beta2/16 at 1.097e-13, the fitted C = 1.1206e-4 and the s = 3.0 crossing 10^15.30, the onsets 4.390e5 / 1.111e6 / 1.860e6, ln(n1n2) = 7.4993e7, and the L7 CRT counts EQUAL at x = 7, 11, 13), so the truth-gap reading of REC is CONFIRMED and the growth half stays DERIVED and red-teamed twice, not proven; what does not survive is the scoping, the exit-window and the price wording: the proposed clause \"at equal levels D1 = D2 = z^s\" is NARROWER than the derivation supports, since over 1643 admissible upper-and-lower level splits the floor beats the decoupled window at every cell by at least 0.644444, so no level split escapes and the sibling rider \"asymmetric levels give 8s1/9 + 8s2/9 with s1 + s2 pinned\" is REFUTED as written (it holds only while both s_i <= 3); the record's z ~ 5.6e31 is u0-specific as well as s-specific (10^31.68 at u0 = 4.2165 against 10^35.36 at u0 = beta2, same s), so replacing it with a beta2-convention 10^15.30 mixes two conventions; the onset correction is right numerically and wrong in wording, since at the record's own s = 2.698721 the exact onset is 1.875e6, LATER than the quoted analytic 1.055e6, not earlier; the escape-(a) correction is right in substance but its \"(A1A2)^2 / (2 n1n2)\" is a LOWER bound on the split's contribution, and the two-sided version through lambda+(n) <= 2^omega(n) reads 10^{-3.004323e+7} against B ~ 10^{10.1}, still no contradiction; and on the L7 side the union-bound argument and the CRT count are CONFIRMED while the coupled vector-sieve price 2(1 + sqrt e) = 5.297442541400 is the FORCED-equal-level constant, the corpus's live figure being K_BF = 5.158064680330, reproduced here to twelve digits at (a, b) = (2.229949, 2.928115), and the sieve-on-holes is the vector sieve only after the Ford-Halberstam substitution step that sift-limit-attack.md 7b(1b) already owns; the \"no third mechanism excluded\" line is honest but incomplete, since weighted sieves with Chen switching are already priced and closed at sift-limit-attack.md 4.2; scope of death gains one row, L7's mechanism 2, which the growth note's own section 10 calls untouched while its sibling calls it dead; no exponent moved. | [redteam-0904-item0.md](history/staging/redteam-0904-item0.md) |\n| 0 | `Q-redteam-0904-sifting-limit` Do recon-0904-sifting-limit-floor.md's five claims survive adversarial re-derivation at the page, and in particular is the 4.26645 cap a method artefact at rung MEASURED? | ANSWERED | Four of five claims survive with corrections and the headline does not. Claim d is REFUTED as stated on four grounds, each sufficient: the level-D LP fixes one omega-profile out of a class Ford defines by a ONE-SIDED inequality, and inside the classical (Omega_2(kappa,L)) budget a legal profile at x = 13 gives a calibrated reading of 5.0113, above beta_2 = 4.26645, which fires the target note's own pre-registered refutation of its floor; the calibration is a 1.72 multiplicative correction fitted at one point and it misses the second proven point beta(1/2) = 1 by 66 to 82 per cent, two and a half times the 28.7 per cent effect the headline rests on; the pooled 3.3152 averages four statistics of which only s* is the definitional analogue of beta, and that route reads 3.4678 here and 3.9487 at x = 43 in the corpus's own data, still climbing at +0.4297 per ln x with the fitted line reaching 4.26645 at x = 104; and 3.3152 is 22.2961 per cent below 4.26645 rather than the 18 per cent the note prints. Claim a is CONFIRMED with a second source off the OCR channel (Ford 2023 p. 37, \"Some authors, e.g. Selberg, refer to 1/beta(kappa) as the sieve limit\"), one page correction owed (Halberstam is p. 117, not p. 116). Claim b is CONFIRMED and strengthened: Brady's class is stated on his p. 10 and IS the axiom-only interval class Face 4 needs, his v_R table is reproduced here, and the identity v_{2d+2} = v_{2d+1} his proof asserts is confirmed at d = 0, 1, 2, so 1.819592 rests on proved facts alone. Claim c is CONFIRMED but demoted: beta(2) >= beta(1) = 2 is two lines from Ford's own one-sided (Omega) plus his own beta(1) >= 2, so the padding construction is correct and unnecessary. Of the eight proposed live-layer changes, two survive verbatim, five need rewording and one is refuted; the new SEARCH-CONVENTIONS row adds exactly three clearing phrases and changes nothing for any existing absence claim. | [redteam-0904-sifting-limit.md](history/staging/redteam-0904-sifting-limit.md) |\n| 0 | `Q-rho-maximal-law` What is the rho maximal law, what is its weakest sufficient form against the crossing, and where does it sit against print? | PARTIAL | The pricing lemma turns any RML(alpha) with alpha < beta_2 = 4.26645 into G2(z#) << z^alpha ln^2 z, every step proven except the law itself; the sufficiency curve lambda_max falls from 2.31949 at z = 13 to 1.78446 at z = 47, the weakest measured point, and whether it ever reaches zero is model-dependent and undecided, so nothing here is an unconditional exponent. | [rho-exact-z31-01.md](history/staging/rho-exact-z31-01.md), [rho-maximal-law.md](history/staging/rho-maximal-law.md) |\n| 0 | `Q-rho-sup-z41-0830` What is the exact sup over all positions of the centred sawtooth remainder at z = 41 (period 7.42e12), and of R_H at H = z^3 and z^4, and what does the eighth exact point do to the truth's slope and to the u0 = 3 form of REC(s, u0)? | PARTIAL | Not run in full; the exact z = 41 point prices at 4.8 to 5.4 h on this machine from measured per-position costs (naive 69.8, table 23.5, table-plus-windows 36.6 ns per worker; fleet 4.52 ns of wall at z = 41), above the 3.5 h rule, so what exists is a new engine (a PROVEN reflection that halves the walk, a P(29) periodic table worth 2.9x, per-block exact re-seeding) reproducing every cited exact sup at z = 13..37 digit for digit, sup\\|R_H\\| at H = z^3 walked for the first time at z = 19..37 (F3 = 26.4 to 35.9, the factor 2 was 1.5 to 2.2x loose), a sealed pre-registration (forecast sup 60.07, band [47.25, 76.37], kill needs sup\\|R_H3\\| >= 1807.417), one pilot twelfth giving lower bounds sup\\|rho~\\|(41) >= 70.651250 and sup\\|R_H3\\| >= 64.416936, and an eleven-segment plan for the box; no bound-and-prune exists on three routes; nothing asymptotic can move whatever the point turns out to be. | [measure-0830-rho-sup-z41.md](history/staging/measure-0830-rho-sup-z41.md) |\n| 0 | `Q-rho2-bound` Is there an analytic upper bound on (z) from its proven closed form, and does it matter for RML? | ANSWERED | Already proven 2026-08-21 (attack-rhoms-01); the constant is now named, 170.88 z^{2s} ln^8 z; Chebyshev off any second moment diverges from RML (the true C_L = ms = 1 floor clears at z = 41 and 43 and misses by 2.155x at z = 47), so it is an ingredient, not a route. | [attack-rhoms-01.md](history/staging/attack-rhoms-01.md), [rho2-analytic-bound.md](history/staging/rho2-analytic-bound.md) |\n| 0 | `Q-rml-proof-0829n` Starting from Lemma V's PROVEN mean-square bound, what is the exact chain of implications to a two-class exponent below beta_2 = 4.26645, which single arrow is open, and does it close? | PARTIAL | The chain from the proven mean square to an exponent below beta_2 has exactly one open arrow, the mean-square-to-sup recovery REC(s,u0): sup_x \\|R_H(x)\\| <= z^{u0/2 - eps} rms(R_H) at H = z^{u0} for one fixed u0 in (2, beta_2); every attempt here dies at a named place (term-by-term accounting is capped at 2s + o(1), the trivial-level exponent; position-union pays e^{theta(z)/q}; window smoothing and the P(y) lever move polylog only; the smooth-profile form is closed on hypothesis and priced above beta_2), the arrow is measured TRUE with certified margin z^1.72 to z^1.92 at u0 = 4 over z = 19..37 and the truth's slope 2.766 +/- 0.212 sits 5.8 se under u0 = 4 on seven points under one octave, so the failure is a proof gap at every computable z and undecided asymptotically; no exponent moved. | [attack-0829n-rml-proof.md](history/staging/attack-0829n-rml-proof.md) |\n| 0 | `Q-tail-deficit-model` Does the exact compound-geometric thinning of the previous level's own gap histogram, with the qualifying-gap constraint g = 0, +/-2 (mod q) carried through the ladder, predict the tile's tail-deficit factor G2/(mbar ln D) at zero parameters, and does that make object-models-read-0829's D5 residue-level classification constructive rather than inferred? | ANSWERED | NO. The zero-parameter CRT-constrained thinning predicts a tail HEAVIER than the tile's at every level, delta_M2 = 0.6461, 0.6746, 0.7021 against a measured 0.4577, 0.4463, 0.4791, so the pre-registered \"classification was wrong\" arm fires and D5 returns to OPEN with C3's constructive route REFUTED; the constraint is informative but small, carrying 4.0 to 10.0 percent of ln(null/truth) at the far tail, or 7.7 to 14.4 percent read against a rate-matched control, against the grain's ORDER at 39.4 to 54.6 percent at the deepest abscissa each level resolves; at shallow abscissae the ordering reverses (B 8.7 against C 4.8 at x = 23, u = 4), which the exact operator on a shuffled true grain confirms without any model (maximum 276, 354, 426 against the true 204, 258, 348), and the order effect is now localised (sections 7 and 9): a first-order Markov word from the true adjacent-pair statistics carries 32.3, 34.2 and 60.8 percent of ln(shuffled/truth) at the deepest abscissa of the three folds, and a SECOND-order word from the true adjacent-triple statistics carries 97.3, 97.9 and 102.6 percent there, r = 0.972, 1.061 and 1.000 on draw-mean maxima over 40, 20 and 10 draws, so the folded tile's tail is MEASURED to be a function of the previous grain's three-consecutive-gap statistics and of essentially nothing longer-range; that surrogate is seeded on the true previous grain at one fold, is not iterated, predicts no delta and does not revive C3, and the 348.0 sd 0.0 at 29->31 rests on ten draws. | [measure-tail-deficit-0829.md](history/staging/measure-tail-deficit-0829.md) |\n| 0 | `Q-theta-ladder` Does the all-positions exponent theta turn over below 2, converge to exactly 2, or settle above 2? | CLOSED | The answer is above 2 with an amendment, and the route is retired regardless: theta does not turn over (it dips and comes back, need/z^2 = 0.3550 to 0.6119 at z = 13..31 and rising from 31), convergence to exactly 2 is not supported (e = 1 rejected at 95% in all four fits), point estimates run 2.37 to 2.99 without settling, and OUTCOMES.md records the certificate route as RETIRED because the sharp maximal law it needs is itself TPC-implying. | [theta-ladder.md](theta-ladder.md) |\n| 0 | `Q-usup-convention-0830` Are lemmaV-sup-extension.md's u_sup and u_sat and attack-0829n-rml-proof.md sec.4.1's CAP(z) statements about the same object (level D = z^s at s = 3.0, same moduli, weights and normalisation), so that the sec.4.1 correction in passing applies, or does a level-convention mismatch void it? | ANSWERED | SAME object, VERIFIED at the code: both notes build the lattice through buildTerms(z, z^3), the modulus sets and Vabs(e) agree to 6.9e-18 at z = 13..23 and the count convention matches at all ten levels, the cited Ssat and u_sat reproduce on the RML lattice to every printed digit at z = 13..19 and the full-level alternative does not (19.544 against 19.602 at z = 13, 45.827 against 50.314 at z = 17); so the PROVEN cap Ssat <= CAP <= z^{2s+o(1)} applies and refutes lemmaV-sup-extension.md's asymptotic prose at lines 485-486 and 508-516 (2^pi(z) moduli, a C^pi(z) theorem as the reachable end), while sec.4.1's attribution sentence overreaches by calling that note's \"reading\" a shape it lists as one of two indistinguishable fits and flags as its likeliest error; neither ledger verdict line changes. | [verify-0830-usup-convention.md](history/staging/verify-0830-usup-convention.md) |\n| 1d | `Q-applied-0828-closures` Were the closures red team's corrections applied to the five HELD notes? | ANSWERED | Applied. 47 prose corrections across the five notes; c2prime-refit-22 drops from CLOSED to PARTIAL and three ledger verdicts are rewritten; four producer READINGS blocks and eleven live-document lines carry the same defects and are listed here uncorrected, behind the fence. | [applied-0828-closures.md](history/staging/applied-0828-closures.md) |\n| 1d | `Q-delta-reader-0830` Can any instrument read the sign of the log-power correction delta in G2 ~ c n^beta (ln n)^delta at reach 79, passing the one-class control first? | ANSWERED | No. Six readers were pre-registered against stepped synthetic laws and the one-class control; the four power-chain readers (the diagonal meter among them) are killed by the P(n) stepping alone, the joint fit is calibrated but reads the control's finite-reach delta NEGATIVE (-0.38 +- 0.14 at 64 terms) where its conjectured asymptotic delta is +2, and the one reader that passes does so by importing the exponent, its sign being the sign of (1.777 - beta_hat). delta's sign for G2 is unreadable at reach 79 by any instrument tried, and the control shows a finite-reach reading would not carry the asymptotic sign anyway; item 1d's all-bases form stays unattackable. | [measure-0830-delta-reader.md](history/staging/measure-0830-delta-reader.md) |\n| 1d | `Q-fekete-1d-defect47` Does the bounded-defect Fekete route survive a probe at 47#? | PARTIAL | The defect at 47# is ordinary and the 47# enumeration cannot move item 1d's TPC threshold whatever value it returns; the bounded-defect Fekete lemma is stated exactly and proved with every hypothesis except the candidate itself discharged, so the route reduces to one named inequality and is neither closed nor open beyond that. | [fekete-1d.md](history/staging/fekete-1d.md) |\n| 1d | `Q-g2-falls-decision-rule` Does G2(x#)/x^2 fall, and what pre-registered decision rule would settle it? | PARTIAL | The instrument is calibrated and sharp, resolving the one-class control at 11.6 sigma on the same nine points where it returns 0.9 sigma on G2, and the honest band on the slope is +/-0.117 at nine terms, so the rule is registered; but the power analysis puts separation at x = 53 only if the falling law is x ln^2 x and at x = 151 if it is x ln^3 x, so phase 1's extra terms settle nothing and no reachable exact ladder does either. | [phase1-T3prep-decision-rule.md](history/staging/phase1-T3prep-decision-rule.md) |\n| 1d | `Q-hsub-reductions` Can (H-sub) be proven or refuted, and what does the fold machinery reach? | PARTIAL | Neither, but the statement is smaller: three reductions of the hypothesis are proven, the lemma consuming only power pairs so that it weakens to (H-sub-pow) with the conclusion unchanged, the fold machinery's inability to reach any of them is made exact, and the hunt over all 111 reachable pairs found no drifting family and no counterexample. | [attack-hsub-01.md](history/staging/attack-hsub-01.md) |\n| 1d | `Q-hsubpow-K` Can (H-sub-pow) be proven with an explicit K in the legal zone? | CLOSED | NO by three mechanisms: the CRT lift (K* >= pi(y') - pi(y) diverges), anchored caps (diverge at every base), Iwaniec at two classes (it IS beta2-note); TODO 1d's legal zone is mis-stated, the trusted zone is [1.3946, 11.3568). | [hsubpow-explicit-K.md](history/staging/hsubpow-explicit-K.md) |\n| 1d | `Q-hsubpow-K-0829n` Can (H-sub-pow) be proven with an explicit K inside the trusted legal zone [1.3946, 11.3568) by a mechanism the 2026-08-28 pass did not close? | OPEN | No K is proven at any base; the single open inequality is the uniform-in-k ratio cap G(b^(k+1))/G(b^k) <= e^K G(b), which is a proof gap at a fixed base and a possible truth gap across bases, since for any law G ~ c n^beta (ln n)^delta the all-bases hypothesis holds with finite K if and only if delta >= 0. | [attack-0829n-hsubpow-K.md](history/staging/attack-0829n-hsubpow-K.md) |\n| 1d | `Q-import-interp` Does the Bayati-Gamarnik-Tetali interpolation method reach the G2 Fekete object? | ANSWERED | No: BGT closes on hypothesis H1, since the constraint count never splits (pi(st) - pi(s) - pi(t) is zero at no pair with st >= 25, minimum 2, maximum 12, VERIFIED on 104 pairs), and the one coordinate that does split has limit +infinity; the row lands as WALL-ADDRESS plus the identity that the submultiplicativity defect is the Overshoot slack. | [import-interp.md](history/staging/import-interp.md) |\n| 1d | `Q-redteam-0828-closures` Do the five 2026-08-28 closure notes stand, and do they stand for the reasons they give? | ANSWERED | Three stand on corrected sentences (1d, Z5b, 0); one stands in half (10, the 1.074 counter-claim does not); one does not stand for its stated reason (1c, closed on a narrower window than the prior work's significant one, and its CLOSED has overwritten a PARTIAL in the generated index). | [redteam-0828-closures.md](history/staging/redteam-0828-closures.md) |\n| 1d | `Q-redteam-0830-fekete` Do the load-bearing claims of attack-0829n-hsubpow-K.md (the sign lemma, the K -> beta_bound map, the zone, the TODO.md base-16 correction) and measure-0830-delta-reader.md (the six pre-registered readers, their kill gates, the joint fit's calibration, the RB circularity argument) survive independent re-derivation, and are the two live-doc corrections they caused the right sentences? | ANSWERED | Every number in both notes reproduces on independently written code (all 14 cells of the beta_bound / K_min table, all 48 cells of the reader-calibration table, all 24 figures of the real-control table, the margins, the sups, the 5.607 correction, the 10-of-81 count), and the sign lemma is re-proven as stated for continuous exact laws. Two claims are WEAKENED and one live-doc sentence needs a qualifier: the sign lemma's constant 1.0597 loses 43% of its value if the base floor moves from 2 to 3 and its whole shape fails for the stepped law that a primorial ladder must present, where beta re-enters the sup and a delta = 0 law already needs K = 1.37 at beta = 2, inside the trap; and TODO.md 1d's \"the sign of delta is unreadable for G2 at reach 79 by any instrument tried\" is wrong on the letter, since reader RB passed all four pre-registered criteria and read delta_hat = +0.80 +- 0.09, the note voiding it post-hoc on a circularity the pre-registration flagged but did not make a kill criterion; the honest sentence adds \"non-circular\". No claim in either note is refuted. | [redteam-0830-fekete.md](history/staging/redteam-0830-fekete.md) |\n| 1d | `Q-y2-recompute` Does the sixteen-level Y2 ladder reproduce on the corrected instrument, and does Y2/x^2 fall? | ANSWERED | The ladder reproduces exactly, sixteen of sixteen from sixteen independent processes, digit for digit against the published 2026-08-18 ladder, and Y2/x^2 falls at 15 of the 15 consecutive steps; this does NOT show that G2(x#)/x^2 falls, since the shortfall Y2/(G2-1) is unmeasured above x = 79. | [y2-recompute.md](history/staging/y2-recompute.md) |\n| 0b | `Q-gate-multiplies` Does \"the gate multiplies\" close the entire u-frame recursion branch, or only the merge chain? | ANSWERED | Only the merge chain and one relative of it: the no-fixed-point argument does not reach TODO 0b at all, though 0b is wrong as stated for an unrelated reason and the error is a factor of ln u; what survives is the copy theorem for maxsum (VERIFIED 40 of 40), which needs a residue-deleted maxsum bound that does not pass through a kill count. | [gate-multiplies.md](gate-multiplies.md) |\n| 0b | `Q-import-thinning` Do point-process thinning and interval coalescence give H'' a null model? | ANSWERED | Yes and exactly: the fold recursion is an exactly solvable thinning whose gap pgf transforms by a Mobius map fixing 0 and 1, so under independent thinning the gap word of T_x is exactly geometric with mean mbar/6 at every rung with no error term, which turns both of the record's fitted parameters into derived ones. | [import-thinning.md](history/staging/import-thinning.md) |\n| 0c | `Q-0c-holesweep` At a fold, how does maxsum degrade copy by copy as the two deleted residues sweep through their possible positions? | ANSWERED | The full curve is on disk for nine folds (5->7 through 31->37, every copy) and it is a PALINDROME: the Mirror-Sweep Lemma is proven here and verified exhaustively, giving the fingerprint any kill-count-free bound must reproduce, but no such bound was produced by this pass. | [attack-0c-holesweep.md](history/staging/attack-0c-holesweep.md) |\n| 0c | `Q-0c0e-level-selection` Can a residue-deleted maxsum bound that only has to hold at levels of the argument's choosing carry the weak Zone Postulate? | CLOSED | CLOSED, and the mechanism has a name: the level-selection licence is worth 0.3187 nats once and is falling, no selection rule beats its own null by anything worth having across T_11..T_29, and the BV/EH sweep supplies no moduli-selecting statement to spend. | [attack-0c0e-level-selection.md](history/staging/attack-0c0e-level-selection.md) |\n| 0c | `Q-foldL-maxsum-direct` Can maxsum_m(T_x), plain and residue-deleted, be bounded above by an argument that never counts kills? | CLOSED | The brief's premise contradicts the corpus's own Bridge Floor (maxsum_k >= maxsum_1 = G2(T_x) for every k), so the bridge cannot convert far enough however good the bound is; all four routes close (R1 partial then closed, R2 tautological and quantitative, R3 busts the budget), and the order-m object is in print, so item 0c's absence sentence has to change. | [attack-foldL-05-maxsum-direct.md](history/staging/attack-foldL-05-maxsum-direct.md) |\n| 0c | `Q-import-chaining` Can Dudley's entropy bound or Talagrand's generic chaining beat the union bound over positions on the maxsum law? | CLOSED | No, and the reason is geometric rather than probabilistic: the entropy integral of the true increment metric is already 1.054-1.099 times rms*sqrt(2 lnW) at its lower branch and 1.484-1.545 at its upper, flat across z = 13..23, so chaining's ceiling with a perfect universal constant sits below the union bound's floor; the honest constant-carrying chain measures 2.878 to 3.027. | [import-chaining.md](history/staging/import-chaining.md) |\n| X | `Q-verify-cofactor-convolution` Does the cofactor-convolution identity at the anchor hold, and does it test the joint law? | ANSWERED | RESTATED with corrections: the arithmetic is right, an independent re-derivation reproducing every figure to the digit at all five levels and confirming the asserted closed-form step, but the framing does not survive, since X is pinned by the identity X = M - Nbar + n_0 so the whole comparison collapses to one cell and the joint is measurably not a product elsewhere; the model error is 0.52 per cent, not 0.050. | [verify-cofactor-convolution.md](history/staging/verify-cofactor-convolution.md) |\n| X | `Q-xchan-at29-prereg` Does the joint-deficit closed form survive a blind test at @29? | OPEN | Pre-registration only, committed alone before any producer existed: the statistic, the predictions adopted verbatim from the record, two acceptance bands, the validation gate the instrument must clear before any @29 number is reported, and what each verdict does to TODO item X. | [xchan-at29-prereg.md](history/staging/xchan-at29-prereg.md) |\n| X | `Q-xchan-at37-offset` Does any registered offset-correction candidate for the ~3.8 law survive at @37? | ANSWERED | Sealed and committed alone before any @37 census of any kind existed, fixing the candidates, the sigma model and its projection band, the scoring rule and what each verdict does to item X's offset clause; scored in xchan-at37-score.md, where every registered candidate is killed and the number survives an independent recount. | [xchan-at37-offset-prereg.md](history/staging/xchan-at37-offset-prereg.md) |\n| X | `Q-xchan-at37-score` What does the @37 census say about the sealed X-channel offset pre-registration? | ANSWERED | The census measured 1 - J = 0.020823, below every registered prediction, so scored exactly as registered every one of the seven candidates dies at \\|z\\| = 104 to 122 and the survivor set is EMPTY, an outcome the sealed prereg has no consequence clause for; the number itself survives an independent recount on a different marking scheme, and the offset question is replaced by a larger one. | [xchan-at37-score.md](history/staging/xchan-at37-score.md) |\n| X | `Q-xchannel-at23` What is the fifth point of the X-channel statistic, at @23, and does the monotone rise hold? | PARTIAL | The fifth point is +0.2658 at @23, the rise holds at five levels and is larger than the four-point trend predicted, and the deficit has migrated into m >= 3, whose share of the X-gap runs 10.5, 23.5 and 81.1 per cent at @17, @19 and @23; the constant itself is not derived here. | [xchannel-at23.md](history/staging/xchannel-at23.md) |\n| X | `Q-xchannel-closedform` Does the zero-parameter closed form for the joint deficit survive its two blind tests at @29 and @31? | PARTIAL | 1 - J = 4S2 is the only one of the three pre-registered laws left standing and it is not exact: a clean hit at @29, z = -0.90, and 4.93 sigma low at @31 where sigma_J falls to 0.000024, with the two rivals dying at 6.35 and 20.81 sigma; both blind levels agree on a stable relative offset of about half a percent, -0.41% and -0.48%, whose absolute residuals match to 2%. | [xchan-at29.md](history/staging/xchan-at29.md) |\n| X | `Q-xchannel-offset` What produces the ~3.8 law's -0.45% offset, and which candidate correction survives? | PARTIAL | The best available description is a finite-level correction of S3's size and sign - 1 - J = 4S2 - S3 scores z = +0.04 at @29 and -0.41 at @31 where the standing law scores -0.43 and -2.32 - but it is POST HOC; two candidate corrections and one shape are refuted on existing data, the sigma behind the @31 detection is calibrated on seven known-truth controls, and the deciding blind test at @37 is pre-registered and committed alone. | [item-x-offset.md](history/staging/item-x-offset.md) |\n| X | `Q-xchannel-triples` Does the joint deficit deepen or decay as the mixed super-W triple census is extended past @17? | ANSWERED | It decays: J = 0.9025, 0.9599, 0.9659 at @17, @19, @23, so the deficit runs 9.7%, 4.0%, 3.4%, both new points land inside the pre-registered DECAYING FASTER THAN FORCED band, and the DEEPENING instinct is refuted by 3.50 sigma between @17 and @19 and a further 1.68 sigma to @23. | [xchannel-triples.md](history/staging/xchannel-triples.md) |\n| A | `Q-anchored-ladder` Does the unified anchored ladder generalize past @11 and @13, and what does cheap K buy at each level? | ANSWERED | It generalizes verbatim to @17 and @19, and the K = 0 unified floor is DEAD there (vacuous at -1262 and -55865); what survives and grows is the family lead at the classic certification depth K*, margin +2, +5, +26, +110 across @11 to @19. | [attack-anchored-02.md](history/staging/attack-anchored-02.md) |\n| A | `Q-anchored-ladder-17-0830` At @17 and above, at which depth K does the composed (unified) anchored floor first turn positive and at which K does it reach the truth, and does the depth-cost curve K_first_positive(level), K_truth(level) move with level? | ANSWERED | At @17/@19 the answer was already on record (Q-anchored-ladder: positive at K = 2/10, truth at ascending K = 109/410, minimal pool 88/350) and is re-derived here as the custody gate; the new level @23 ran under a sealed 14-of-14 forecast: positive first at K = 27 = K* (PROVEN from the record before the run), truth at ascending K = 1732 of 1739, minimal pool 1543 = 0.887 of the scour (PROVEN lower bound meets VERIFIED upper bound); across five levels positivity costs exactly K* while the truth fraction rises 0.40, 0.62, 0.73, 0.81, 0.89, a MEASURED finite table with no law in x and no exponent touched. | [attack-0830-anchored-ladder-17.md](history/staging/attack-0830-anchored-ladder-17.md) |\n| A | `Q-anchored-unify` Can the three anchored mechanics be unified at @11, and what floor family does the unified cap certify? | ANSWERED | MEASURED and PROVEN-here but HELD: extending the staircase's cofactor injection to m = 1 folds the self-strike allowance into the count, the unified-cap lemma gives fresh <= capU_K <= cap_K at every depth and prime at @11 and @13, and the floor family runs 36 to 45; the x = 37 secondary is a clean negative. | [attack-anchored-01.md](history/staging/attack-anchored-01.md) |\n| A | `Q-history-dial` Does revealing the scour primes' classes concentrate the floor on a bounded set of strike locations, and what is the history dial worth? | ANSWERED | Disconfirming: the class axis is linear at 2.90 survivors per prime read, so no bounded set of strike locations carries the floor, and the touch-count ceiling is proven and tight; deliverable (1) of the brief already existed as canonical state (the forcing ladder 16 -> 45 at @11), so nothing new was bought there. | [attack-history-dial.md](history/staging/attack-history-dial.md) |\n| A | `Q-redteam-0830-engine` Do attack-0830-buchstab-deep.md, attack-0830-comb-tail.md and attack-0830-anchored-ladder-17.md survive an adversarial re-derivation on code that shares nothing with their producers, and does REFUTED row 97 stand as worded? | ANSWERED | Every headline number in the three notes reproduces exactly on independent code (26 of 26 comparisons, including F2/f2 = 3.6682 at sigma = 4.7088 by a near-analytic route, the 0-of-1512930 pair count, the tail defect at five levels, the full 419328-term Legendre split, and the whole @23 anchored ladder including K* = 27, floorU = 5364, truth 597475 and the minimal pool 1543 proven exact); one load-bearing DERIVATION is REFUTED — \"\\|B - 1\\| >= c forces sigma(q) < u_c - 1\" fails at @23 on 21134 (q, K) pairs for the engine's model B at c = 0.1 and on 33875 high-count pairs for the measured B at c = 0.01, because the step bounds every u in a weighted average by the largest one and treats the denominator as e^-gamma, so the \"level-free\" clause of buchstab-deep section 4 and REFUTED row 97's \"\\|B - 1\\| >= 0.01 forces sigma < 1.798\" must be re-graded from DERIVED to MEASURED-at-@23 (where the conclusion survives: the worst sigma is 2.159, still far below beta2 = 4.26645); row 97's \"the best sigma any level has\" is separately WEAKENED, since @29's head sigma is 5.5785 with width 1.5325 and @97's is 17.1422 with width 1.0000; comb-tail derives its main-term factor from prod (1 - 1/q_i) and measures it against prod (1 - 1/(q_i - 1)), which is what natal-cap-28 line 283 computes against its own line 20 and against certificate-engine.md section 1, a 0.9% difference at @23 that moves nothing certified; comb-tail also misquotes its own cited source as F1 in [1.5, 4] where that source says [1.5, 3] and says there is no bound at all at s < 1; four smaller slips are listed and none opens or closes a route. | [redteam-0830-engine.md](history/staging/redteam-0830-engine.md) |\n| 1 | `Q-at43-bigint-0830` Can the K-30 natal march be carried past its 2^53 ceiling to @43 exactly, and is the @43 point worth its cost? | PARTIAL | The engine side is done and VERIFIED (five paths promoted to BigInt, @7..@37 and a 0.248% slice of @41 reproduced digit for digit); the @43 point is NOT run, because the measured extrapolation is 579 h of eight cores on this machine (a 43x tile times a 2.23x per-cell cost, against the brief's ~13x), which is a weeks-class box job whose only payoff is a sixth point on a curve with no consumer; the forecast 0.8393 raw / 0.8399 persisted stands unscored. | [engine-0830-at43-bigint.md](history/staging/engine-0830-at43-bigint.md) |\n| 1 | `Q-redteam-0830-imports` Do the six lighter notes of 2026-08-30 (varE-identification, smooth-aps, rec-killrun, skeleton-door, at43-bigint, coherence) survive an adversarial pass on independent code and at the source pages, and which of their load-bearing claims are wrong? | ANSWERED | No note is refuted whole and no exponent moves. Every decisive measurement reproduces on code sharing nothing with the producers: the varE mixed remainder and its 0.844 / 0.826 share below 2L and 94.4 / 98.7 percent unbalanced share at x = 13, 17; the Buchstab deviations and all three crossing levels log10 z* = 33.53, 68.80, 33.62; the skeleton door's 9.2 / -0.9 percent, G30_agg 0.1113 / 0.1011 and the 0.0102 cap at @13, @17, by a path that also verifies cap-36 Theorem A; the @43 parity exactness on all 26,157,448 scour classes with 0 above 2^54 and a constructed @47 counterexample; and the five recon-0828 verdict counts summing to 75, which REFUTES rec-killrun's brief-error clause and confirms the orchestrator's rider. Seven claims WEAKEN. The load-bearing one is smooth-aps's \"the weight is unmet in print\": Harper's J. London Math. Soc. 112 (2025) e70293 proves a Barban-Davenport-Halberstam asymptotic for an ARBITRARY sequence on exactly that dyadic all-classes object, and his own footnote says the missing max over x' can be incorporated, so the NEAREST citation changes even though the step still does not close (the 2025 range is sqrt(2x) < Q <= x and its Theorem 2 route bars sieved sets by name). Also: Henriot Corollary 2 applies, but to the ERRATUM's statement and under two hypotheses the note leaves unstated (eps < alpha/600, not \"every eps\"; F at prime powers); varE section 4's display (*) carries an unfilled placeholder and is vacuous by a factor L/d; the skeleton door's \"would tighten\" is wrong in sign at @17, where the closable block is negative and removing it RAISES G30_agg to 0.1020; Halasz-Montgomery is not applicable for the reason given (it is an abstract inner-product inequality); FGKT's heuristic quote is on p. 4, not p. 3; and the \"~13x had no source\" is a corpus miss, the retired G2-walk figure at CHANGELOG.md:5716. Not TPC-strength; nothing here opens a route. | [redteam-0830-imports.md](history/staging/redteam-0830-imports.md) |\n| Z7 | `Q-zonegap-03-score` Do the ten predictions sealed in zonegap-03-prereg.md score against the stage-3 sweep at X = 1e12? | ANSWERED | All ten sealed rows score HIT and none miss, after a second pass added a band-edge argument to zonegap-01.js and re-ran the decade so the two rows that named an unprinted band could be scored; the four Group T hits are a custody promotion that follows from CUSTODY 1 and 3 passing, one sub-clause of T5 (the full-decade sd) stays unprinted, and the sweep still needs a DERIVED engine because zonegap-01.js's inlined 41-record ladder makes it exit at 1e12. | [zonegap-03-score.md](history/staging/zonegap-03-score.md) |\n| 2 | `Q-var41` What does the stable law predict for Var(41), and what can the tenth Var/E point pin? | OPEN | Pre-registration only, sealed and committed alone before any Var(41) engine exists: it freezes the prediction, a band taken from the law's own residuals at z <= 37, the derived z(41) prediction, and the honest statement that one more point cannot separate a limit from a drift. | [var41-prereg.md](history/staging/var41-prereg.md) |\n| 8 | `Q-applied-0828-engine` Were the engine red team's corrections applied to the five HELD notes? | ANSWERED | Applied, 37 edits across the five notes: two numbers corrected (first s >= 10.82 at x = 263 not 239, first non-empty kappa=1 level @37 not @53), two ledger verdict lines and one note title rewritten, one proof step given BV's max-over-y form, the one-class freshness factor restated on P^-(m) >= q so it holds at every scour prime, and two mislabelled columns renamed; six corrections that land on live documents are collected here unapplied, and no producer was touched. | [applied-0828-engine.md](history/staging/applied-0828-engine.md) |\n| 8 | `Q-applied-0828-live` Which engine and head red-team corrections reached the live layer? | ANSWERED | Nineteen edits across four documents: the glossary's cap_K dimension count with its mandatory sifted-versus-assumed clause plus three new entries (twin opener, head, tail); the survey's one-class density factor, S1's restored pi(q-1) terms and r in N_x hypothesis, the dropped Chen absorption, Theorem C with its emptiness, dimension 1 for q_i < q, the comb-conditioning qualifier and the inverted shallow-band bullet; the certificate engine's equidistribution status cell, the q = T^{o(1)} second hypothesis and the Comb Discrepancy scope cell; and the census's R as the continuum functional with its three comparators; paper/staircase-note.md, TODO.md and QUESTIONS.md are owed and untouched. | [applied-0828-live.md](history/staging/applied-0828-live.md) |\n| 8 | `Q-buchstab-deep-0830` Can the certificate engine's deep-ladder Buchstab transfer be proven at dimension 2 at the depths the run levels reach by a route that does not go through the fundamental lemma at s >= 22.06 or the constant-free DH Theorem 9.1 (TODO 8a)? | CLOSED | NO by any instrument this corpus can cite, and the non-closing step is structural, not a constant: the transfer is an asymptotic for a ratio of two dimension-2 sifting functions at one sifting depth z = q, so a sieve bracket [f2, F2] survives undivided in the ratio (width 3.67 at the best sigma any level has, 4.71 at the @23 head); a correction \\|B - 1\\| >= 1% can only occur at sigma(q) < 1.8 where f2 = 0 and F2 >= 7.85, while a lower bound exists only at sigma > 4.266 where \\|B - 1\\| < 1e-5; the Buchstab identity iterated once is already negative at K = 1 for q = 37 at @23 and is the sieve itself when iterated fully; Jurkat-Richert fails Omega(1) as stated; in the certificate's legal direction F2 beats the trivial cap_K <= cap2 at 0 of 1512930 (q, K) pairs at @23, so the sharp-sieve floor is -1733138 at every K; the transfer stays HEURISTIC, what would move it is a kappa = 2 asymptotic in the open band 2 < sigma < 4.266, and part (b) is scoped and graded, not attacked. | [attack-0830-buchstab-deep.md](history/staging/attack-0830-buchstab-deep.md) |\n| 8 | `Q-buchstab-transfer` Is the certificate engine's Buchstab transfer provable in the shallow regime y = T^{o(1)}? | ANSWERED | PROVEN conditional on the fundamental lemma and EMPTY at every run level (needs s >= 22.06; available 4.7 @23, 17.1 @97); the engine's B is a dimension-1 object, the survey's pair-Buchstab is dimension 2, the certificate needs a third. | [thm-buchstab-transfer-shallow.md](history/staging/thm-buchstab-transfer-shallow.md) |\n| 8 | `Q-comb-discrepancy` Can the Comb Discrepancy Lemma's constant 2*3^k be tightened (TODO 8c)? | PARTIAL | The lemma IS the trivial per-block bound; the attained optimum is max G - min G, a computed number per level (slack 29x @23, 54x @29, growing); term-by-term pricing lifts the certified head share to about 28-29 percent, not the 50 hoped for. | [comb-discrepancy-tight.md](history/staging/comb-discrepancy-tight.md) |\n| 8 | `Q-comb-tail-0830` Can TODO item 8's remaining part (b), prime-comb equidistribution in the tail regime, be proven at the depths the certificate needs, or at least measured and priced at the enumerable levels (8b)? | PARTIAL | Not proven and no exponent moves: the tail defect is measured exactly for the first time (signed -0.01% of the tail main term at @23, -0.00% at @29, unsigned 0.60% and 0.25%, smaller than the head's 1.00% in aggregate but 8 to 37% per prime at its worst), the Comb Discrepancy Lemma's method fails in the tail on its MAIN TERM (off by a derived factor e^gamma(1/u - q^(1-u)), measured aggregate 1.14x at @23) before its error term (10^27 x main), the prime-comb Legendre form has 2^(k+2) terms of which the short ones obey the lemma's <1 bound and the long ones are prime-in-AP discrepancies whose term-by-term sum is 11.2% of main at @23 against a signed -0.01%; the all-x form at K = 0 is Theorem C (PROVEN, limit only) and at the certificate's depth K* the non-closing step is a kappa = 1 asymptotic at sifting parameter s < 2, a proof gap on data that show B_tail = 0.998-1.000 at K* through @29; the full-depth half is 8(a), CLOSED. | [attack-0830-comb-tail.md](history/staging/attack-0830-comb-tail.md) |\n| 8 | `Q-redteam-0828-engine` Do the five 2026-08-28 certificate-engine notes survive an adversarial re-derivation, and does the one-class freshness factor correction hold before it reaches a live document? | ANSWERED | The one-class factor holds and is stronger than either note states; four numbers are refuted (first s>=10.82 is x=263 not 239, first non-empty kappa=1 level is @37 not @53, rho*g_max is off by one gap, the @17 certified share and per-term constant carry the wrong convention) and one proof step needs the max-over-y form of BV; the cluster's shared verdict stands. | [redteam-0828-engine.md](history/staging/redteam-0828-engine.md) |\n| 8 | `Q-redteam-0830-engine` Do attack-0830-buchstab-deep.md, attack-0830-comb-tail.md and attack-0830-anchored-ladder-17.md survive an adversarial re-derivation on code that shares nothing with their producers, and does REFUTED row 97 stand as worded? | ANSWERED | Every headline number in the three notes reproduces exactly on independent code (26 of 26 comparisons, including F2/f2 = 3.6682 at sigma = 4.7088 by a near-analytic route, the 0-of-1512930 pair count, the tail defect at five levels, the full 419328-term Legendre split, and the whole @23 anchored ladder including K* = 27, floorU = 5364, truth 597475 and the minimal pool 1543 proven exact); one load-bearing DERIVATION is REFUTED — \"\\|B - 1\\| >= c forces sigma(q) < u_c - 1\" fails at @23 on 21134 (q, K) pairs for the engine's model B at c = 0.1 and on 33875 high-count pairs for the measured B at c = 0.01, because the step bounds every u in a weighted average by the largest one and treats the denominator as e^-gamma, so the \"level-free\" clause of buchstab-deep section 4 and REFUTED row 97's \"\\|B - 1\\| >= 0.01 forces sigma < 1.798\" must be re-graded from DERIVED to MEASURED-at-@23 (where the conclusion survives: the worst sigma is 2.159, still far below beta2 = 4.26645); row 97's \"the best sigma any level has\" is separately WEAKENED, since @29's head sigma is 5.5785 with width 1.5325 and @97's is 17.1422 with width 1.0000; comb-tail derives its main-term factor from prod (1 - 1/q_i) and measures it against prod (1 - 1/(q_i - 1)), which is what natal-cap-28 line 283 computes against its own line 20 and against certificate-engine.md section 1, a 0.9% difference at @23 that moves nothing certified; comb-tail also misquotes its own cited source as F1 in [1.5, 4] where that source says [1.5, 3] and says there is no bound at all at s < 1; four smaller slips are listed and none opens or closes a route. | [redteam-0830-engine.md](history/staging/redteam-0830-engine.md) |\n| 8 | `Q-sharp-sieve-range` Do the sharp sieve functions (Jurkat-Richert, DHR) make the two empty certificate-engine theorems non-empty at run levels? | CLOSED | NO at finite level: DH Thm 9.1 carries no written constant and the crude fundamental lemma is strictly better; as limit statements kappa=1 is first non-empty @37 (a 19.8-wide bracket; @53 is the first narrow one, 1.656) and kappa=2 @23. | [thm-sharp-sieve-range.md](history/staging/thm-sharp-sieve-range.md) |\n| 8 | `Q-u2-engine-depth` Is the certificate engine's operative depth heading to u = 2, and does that make TODO item 8's payout form TPC-strength? | ANSWERED | It is not: the measured operative depth runs u = 4.191 at @17 down to 3.557 at @97 and is still falling, so the brief's conditional does not fire and item 8 is shown neither TPC-strength nor safe; a WALL-ADDRESS in the weakest sense, every figure SCRATCHPAD-GRADE. | [u2-engine-depth.md](history/staging/u2-engine-depth.md) |\n| Z0 | `Q-applied-0828-census` Were the census red team's corrections applied to the three HELD notes? | ANSWERED | Every WEAKENED and REFUTED verdict in redteam-0828-census.md is applied to its target note in the present tense; two producer defects and four live-doc corrections stay open and are listed here. | [applied-0828-census.md](history/staging/applied-0828-census.md) |\n| Z0 | `Q-applied-0828-quadpoint` Were the quadpoint red team's corrections applied to the four HELD notes? | ANSWERED | Applied: the nine REFUTED and WEAKENED verdicts that target the four notes are written into them in present tense, with ledger blocks added; the tenth targets the sealed prereg, outside the fence, and is recorded in attack-quadpoint-02.md instead; three corrections owed by TODO.md Z2 and one by Z0 are listed here and are NOT applied. | [applied-0828-quadpoint.md](history/staging/applied-0828-quadpoint.md) |\n| Z0 | `Q-redteam-0828-census` Do destroyer-census-01.md, stretch-01.md and records-placement-01.md survive an independent re-derivation of every load-bearing claim, and may they leave HELD? | ANSWERED | All three survive on arithmetic, every decisive number reproducing exactly on an independent engine; one sealed reading (records READ-3) is REFUTED as vacuous and four sentences need correcting before integration. | [redteam-0828-census.md](history/staging/redteam-0828-census.md) |\n| Z0 | `Q-redteam-0828-quadpoint` Do the four HELD quadpoint notes of 2026-08-22 survive an adversarial pass: the capture identity's proof line by line, the depth law and its asymptote, the sealed prereg scoring, and the prior-art citations at page level? | ANSWERED | The identity is PROVEN once its window convention is written down and holds at every K of 50 anchors under an independent implementation; four load-bearing sentences are wrong (section 4's gloss on Sum capU, v1 section 2's limit, the prior-art note's 2%, and the note's unconditional y* sentences), and none of the four defects opens or closes a route. | [redteam-0828-quadpoint.md](history/staging/redteam-0828-quadpoint.md) |\n| Z0 | `Q-review-request-0907` Which claims of the 2026-09-07 session should an independent math agent check, in what order, against which records, and what must the review return? | ANSWERED | Reviewed 2026-09-08 by an outside reader (history/reviews-0907/10): ranked claims 1 (conditional on the unread 1974 page), 2, 4, 5, 6 verified within stated scope; claim 3 verified with its quoted tolerances corrected (max deviation 1.8e-4, not 1e-4 or 1e-5); the Corollary 1 slip sharpened (fails for every representative other than 1, and mis-signs even then); ASSUMED items 4 and 6 discharged, 2 partially, 1, 3, 5 not. Twin-prime infinitude and every sufficient signed margin remain OPEN. Ranked below: the §6.2 demotion in the two-class manuscript (a change to a proof's dependency), the D_y identity reading (a change to what a census can show), the Erdős #687 attribution, and the two measurement artifacts. Twin-prime infinitude and every sufficient signed margin remain OPEN. | [review-request-0907.md](review-request-0907.md) |\n| A9 (retired) | `Q-a9-histogram-operator` What is the exact evolution of the gap-size histogram under folding, as a transfer operator on count(d), and what is the tail P(gap >= 2p)? | ANSWERED | The operator is derived and then verified by exact computation at every fold from T5 to T31 and the tail law is computed exactly, but the machinery is a rediscovery of Holt and Rudd arXiv:1408.6002 section 5 and may not be presented as new; only the G2 spacing beyond their \\|s\\| < 2p restriction is ours. | [a3-09-histogram-operator.md](a3-09-histogram-operator.md) |\n| G (retired) | `Q-block-L-first-dead` Is the block-1 counting bound on L the first-dead 62 or the last-alive 111, and does downward closure hold in the block coordinate? | ANSWERED | Downward closure HOLDS in the block coordinate: L <= 62 is sound and recomputed from scratch rather than inherited, L <= 111 is true, implied by 62 and weaker by 1.79x, and the adjudication's supposedly sharper instrument is attack 3's instrument. | [adjudicate-L-bound.md](history/staging/adjudicate-L-bound.md) |\n| 1c (retired) | `Q-c2prime-drift` Why does c2' drift on the 22-term ladder? | PARTIAL | Class-count-blind (attack-c2drift-01 §1); on item 1c's own band x >= 41 the ten-point rate is unresolvable: c2' 0.3745 +/- 0.1411 against c1's 0.3344 +/- 0.2358 on the same primes, a difference of 0.0401 = 0.12 of the control's ten-point window sd. Whether the drift is real is NOT closed here: attack-c2drift-01 measured it at p = 0.0053 on x >= 17 and left (i)' and (ii) alive. exponent-control §2/§4/§6 refit at 22 terms. | [attack-c2drift-01.md](history/staging/attack-c2drift-01.md), [c2prime-refit-22.md](history/staging/c2prime-refit-22.md) |\n| 11 (retired) | `Q-capK-bv` Is the prime-regime cap_K over the full wheel an unconditional asymptotic via the fundamental lemma over Bombieri-Vinogradov? | ANSWERED | PROVEN at short-note grade for q_K = W^{o(1)}, and EMPTY in every computable range (needs s >= 10.82, first cleared at x = 263; the lemma's floor s >= 10 is first reached at x = 239); the deep ladder is untouched. | [thm-capK-bv.md](history/staging/thm-capK-bv.md) |\n| 11b (retired) | `Q-consolidation-wave` Can the five cost-class scripts be bound to their own output, and can U-FRAME's two lost tables be rebuilt? | ANSWERED | All four runnable cost-class embeds are bound and the hand-pasted-tail counter falls 19 to 15 with the gate at TOTAL 0; three needed --force at a cost of 24 tokens out of 211 with no value changed anywhere, and natal-cap-33-overnight.js is a three-run composite that should never have been on the cost list. | [consolidation-wave.md](history/staging/consolidation-wave.md) |\n| 10 (retired) | `Q-excess-chain-c` Does the excess chain's constant c pin to 1.074 via a compressed-tail max correction? | CLOSED | NO: for any FIXED correlation length the derived family tends to 1 and both named corrections lower the maximum, so 1.074 is not derivable from a compressed-tail max correction; c measures 1.05 +/- 0.07 (sd) over x = 17..31, MEASURED, which does not separate 1.05 from 1.074. | [excess-chain-c.md](history/staging/excess-chain-c.md) |\n| 1b (retired) | `Q-greedy-oracle` Is the corrected greedy a G2 oracle? | ANSWERED | ORACLE ESTABLISHED, 13 of 13 exactly-known terms hit with minimum ratio 1.0000, so the greedy RULE is an exact solver where the answer is checkable at x <= 41; the rider matters more, the search budget needed grows 3.31x per additional prime and two objects whose truth reaches further show the same estimator's fidelity DECAYING, so nothing is established for the greedy AS RUN on the ladder. | [greedy-oracle-validation.md](history/staging/greedy-oracle-validation.md), [phase1-T1-greedy-oracle.md](history/staging/phase1-T1-greedy-oracle.md) |\n| 4 (retired) | `Q-import-fracparts` Does Saffari-Vaughan Theorem 10 cover the branches carrying the Skeleton door's mass? | ANSWERED | Prediction HIT: the covered branches carry at most 8 percent of the mass; anchor and a sharper wall-address banked; the THEOREM column is struck, not merely overpriced, because Theorem 10's range condition is x^{6/11+eps} < y, which fails at fixed M; no route; no cap-36 line is upgraded. | [import-fracparts.md](history/staging/import-fracparts.md) |\n| 0e (retired) | `Q-ioslack-survey` Does any instrument in the corpus carry a sharp-level signature that would let the infinitely-often slack be spent? | CLOSED | None does: 23 instruments surveyed and 19 eligible, smallest per-instrument p_FWE = 0.1082 and 0.8865 after Šidák, so dial 4's fourth channel closes and the dial is finished; the second finding is that maxsum_{L+1} is sharp exactly when L = 1. | [ioslack-survey.md](history/staging/ioslack-survey.md) |\n| Z5b (retired) | `Q-lambda-ledger` Does parity information survive in the lambda-weighted fold ledger? | CLOSED | REJECT, against a pre-registered threshold fixed before the producer existed: the lambda multiplier exists and is parity-free, so the ledger is blind at its design point rather than at its margins; two columns are pinned by proof, and nothing here is novel mathematics. | [attack-lambda-ledger.md](history/staging/attack-lambda-ledger.md) |\n| Z5b (retired) | `Q-ledger-forced` Which properties of the fold ledger's columns are FORCED rather than expected? | ANSWERED | Fourteen forced constraints, all identities or wrong-direction bounds; by_new is forced only in the prime gap (its O(1) is MEASURED); nothing forces net from below. | [fold-ledger-forced.md](history/staging/fold-ledger-forced.md) |\n| 11 (retired) | `Q-mod30-tail` Are the BV survey's S1 (mod-30 tail) and S2 (fixed truncation and depth) provable as stated? | ANSWERED | Both PROVEN at short-note grade (no second reader) once S2's freshness factor is corrected to one class per prime, which needs only P^-(m) >= q > q_i and so holds at every scour prime, not only in the prime regime; tail constant 2 ln 2 -> (ln 2)/2; nothing in the head. | [thm-mod30-tail.md](history/staging/thm-mod30-tail.md) |\n| 3 (retired) | `Q-monotonicity-sweep` Does anything in the corpus assume certificate validity is monotone in L? | ANSWERED | One unsound artifact and one invalid inference: attack-beta2-04-loss-budget.js section 6 bisects on a predicate measured not upward-closed and is wrong at 3 of 5 levels, true first-crossings 30/72/132/174/210 against the reported 36/72/144/174/354, and redteam-DP1-certificate.js draws a global minimality conclusion from a two-point local check; corrected, worst-casing certifies within 1.00 to 2.00 of true G2 and the exponent penalty runs 0 to 0.27 and falls with x. | [monotonicity-sweep.md](history/staging/monotonicity-sweep.md) |\n| Z6 (retired) | `Q-records-placement` Do the seven unswept twin-gap records 76-82, above 2^53, land uniformly inside their stretches, or is there square-anchor coupling in record-start placement? | ANSWERED | Sealed alone before the producer existed: the uniform band, the fixed definitions, the registered readings, the calibration abort gate and the riders are all fixed here and no placement fraction, q or stretch boundary was evaluated; scored in records-placement-01.md. | [records-placement-01.md](history/staging/records-placement-01.md), [records-placement-02.md](history/staging/records-placement-02.md), [records-placement-prereg.md](history/staging/records-placement-prereg.md) |\n| 5 (retired) | `Q-shadow-amplitude` Where does the kill shadow's drift amplitude come from, and does the record's counting floor explain it? | MIXED (shadow-amplitude-prereg.md: OPEN; shadow-amplitude.md: PARTIAL) | The finite-y correction is an identity with no free parameter and both measurement routes agree 10 of 10, but the record's own counting floor is four to six times too small and its candidate explanation is the wrong half, so the label stays PARTIALLY EXPLAINED. | [shadow-amplitude-prereg.md](history/staging/shadow-amplitude-prereg.md), [shadow-amplitude.md](history/staging/shadow-amplitude.md) |\n| 5 (retired) | `Q-shadow-prereg` Is the kill shadow's 0.85 the band-average of the Unification-Law survival curve over the post-crystallization window? | ANSWERED | Scored in shadow-buchstab.md: SHAPE-ONLY under the original y >= 997 rule; D3 passes in the reported ten clusters but D1 fails near y = 1000 at 1.49 times the 0.011 tolerance. Later finite-y normalisation and counting-floor corrections do not promote this verdict. The pair-Buchstab identification remains conjectural; the pre-registration body is preserved. | [shadow-prereg.md](history/staging/shadow-prereg.md) |\n| 0d (retired) | `Q-single-alignment` Does the head's single-alignment fold recursion for M(x, x^3) give a proven-shaped handle on the per-fold multiplier that the max-over-alignments recursion lacked? | CLOSED | The answer splits: the multiplier IS different in shape, measured - exactly 1 at 225 of 240 folds from x = 53, total spend 1.09 nats against a tile-shaped ladder's 7.12, measured M(1613) = 2220 against a tile-shaped 3.7e5 - but the route to a growth law through the recursion closes anyway, because the growth is initiated by the window's expansion into fresh ground, a boundary term that is the localized gap problem re-posed and that no fold reaches. | [localized-single-alignment.md](localized-single-alignment.md) |\n| 4 (retired) | `Q-skeleton-decide-0830` Should TODO item 4, \"Skeleton Equidistribution: restate or demote\", be restated around the modulus-W mass or demoted off the board, and what exactly is banked either way? | ANSWERED | DEMOTE. Over every scour prime the branches on which the door is a fixed-modulus question carry 9.2, -0.9, 0.2 and 0.5 percent of the skeleton at @13, @17, @19, @23 (the -10.8 and 5.5 percent on record were 9 and 2 percent subsamples), so the door as named removes at most 0.0102 from a G30_agg whose open part is 0.094 to 0.126; on the open side the phase never wraps and no equidistribution statement remains, only the inequality; nothing on the live board consumes G30_agg < 1/2; no first move exists that is not already ANSWERED. Banked regardless: the Skeleton Collapse Theorem (PROVEN, all x, all q) and G30_agg < 1/2 CERTIFIED at six levels @11..@29. | [decide-0830-skeleton-door.md](history/staging/decide-0830-skeleton-door.md) |\n| 4 (retired) | `Q-skeleton-door` Is the Skeleton Equidistribution Conjecture the blocker on the anchored calm? | ANSWERED | It is not the blocker as far as the door can be seen: Theorem A (Trapezoid Cancellation) is proven for all x, q and branches, @29 is certified as an exact integer inequality at G30_agg = 0.1176 with margin 0.3824 over 7,863 scour primes, and the decay-law shortcut is REFUTED, the six levels being flat within their own spread with every fit made strictly worse by adding @29. | [natal-cap-36-skeleton-door.md](natal-cap-36-skeleton-door.md) |\n| Z3 (retired) | `Q-stretch-structure` What is the stretch S_q = [q^2, q'^2) as an object, where does the Stretch Postulate sit against the Zone Postulate, and does the square anchor's QR structure buy any density advantage? | ANSWERED | The implication chain and the QR kill law re-derive independently arrow by arrow; the anchor's structure is real, deterministic and redistribution-only (exactly two killed offset classes at every prime anchor checked), so there is structure and no advantage, and SP sits strictly above the strong Zone Postulate. | [stretch-01.md](history/staging/stretch-01.md) |\n| 0e (retired) | `Q-verify-monotone-depth` Does the Monotone Depth obstruction, which retires dial 4, survive an adversarial re-derivation? | ANSWERED | NARROWED: every measurement in the record reproduces, several to the last digit, but the mechanism as named is false as a general law and the corpus already holds a counterexample to it; what survives is an accounting of three channels priced in tenths of a nat against a need of 6.700, 8.191 and 9.637 nats, so dial 4 becomes priced slack with one channel open and unpriced. | [verify-monotone-depth.md](history/staging/verify-monotone-depth.md) |\n| 1e (retired) | `Q-wrongdirection-audit` Are any of the ten live targets secretly TPC-strength, before the sessions are spent on them? | ANSWERED | The audit closes no target: two of the ten come out TPC-strength and three more carry a TPC-strength face under a quantifier they are likely to drift into, which is a labelling result and not a reason to drop them; item 1e is settled only because the answer was already on disk, and one route-blocking number in it is hand arithmetic and unstamped. | [attack-wrongdirection-audit.md](history/staging/attack-wrongdirection-audit.md) |\n| Z3 (retired) | `Q-z3-immune` Do the square anchor's immune offset classes (stretch-01 §4) carry a survivor surplus beyond the redistribution-only guardrail, do they attract the floor's first survivor, and does restricting the transplanted caps to them lower K*? | CLOSED | No, on all three. The preregistered surplus test is VOID on its own matched control (a 0.6% ratio-of-means bias present with and without the quadratic point); the preregistered placement test HOLDS at Z = +0.49; the design-free class test reads the guardrail's r/(r-2) over 3.4 million twin openers with a largest per-prime deviation of 0.11% and per-prime Z in -0.48..+1.66; and the restricted certificate, while lowering K* at 61% of anchors, has no certificate at all at 93 of 1219 and is beaten head to head by the QR-BLIND restriction that certifies 1219 of 1219. The refinement is closed. | [attack-z3-immune-01.md](history/staging/attack-z3-immune-01.md) |\n| Z (retired) | `Q-zonegap-reduction` What must be proven for Z2, what already is, and what does the proven head machinery give the zone? | PARTIAL | The Zone Restriction Lemma is proven and verified at five levels - the in-zone twin slots of T_p ARE the in-zone twin primes, so Z2(p) IS the whole-tile gap object restricted to the head window - the three-slack chain is graded honestly, the transfer inventory is listed item by item, and the deep end is proven an onset desert over 24 levels; section 5 lists what was not reached. | [zonegap-02-reduction.md](history/staging/zonegap-02-reduction.md) |\n\n## 2. Every question, by id\n\n| id | status | question | verdict | TODO | records |\n|---|---|---|---|---|---|\n| `Q-0c-holesweep` | ANSWERED | At a fold, how does maxsum degrade copy by copy as the two deleted residues sweep through their possible positions? | The full curve is on disk for nine folds (5->7 through 31->37, every copy) and it is a PALINDROME: the Mirror-Sweep Lemma is proven here and verified exhaustively, giving the fingerprint any kill-count-free bound must reproduce, but no such bound was produced by this pass. | 0c | [attack-0c-holesweep.md](history/staging/attack-0c-holesweep.md) |\n| `Q-0c0e-level-selection` | CLOSED | Can a residue-deleted maxsum bound that only has to hold at levels of the argument's choosing carry the weak Zone Postulate? | CLOSED, and the mechanism has a name: the level-selection licence is worth 0.3187 nats once and is falling, no selection rule beats its own null by anything worth having across T_11..T_29, and the BV/EH sweep supplies no moduli-selecting statement to spend. | 0c, 0e (retired) | [attack-0c0e-level-selection.md](history/staging/attack-0c0e-level-selection.md) |\n| `Q-DP1-mechanism` | ANSWERED | Does anything consume two-point data about a sifted set? | One consumer exists and is priced: the sharp degree-2 Boole-Frechet bound on empty windows is unconditional and concludes at every window position, but it over-certifies G2 with a log-log slope of +5.20 against x over x = 5..19, so the price of stopping at two-point data grows as a power of x rather than as a constant. | none | [attack-DP1-mechanism.md](history/staging/attack-DP1-mechanism.md) |\n| `Q-L-own-law` | ANSWERED | Does L, the adjacent-kill run length, have a law of its own level by level, and which abscissa is it linear in? | L has no law of its own: (L+1)*mbar(T_v) is approximately G2(y#) at every tile and top prime where both sides are exact, spread 1.0000 to 1.0838 over four tiles at y >= 23 with mean 1.0198, so the residue is quantisation, the abscissa question returns a calibrated negative, and the distance to 529 is priced rather than closed. | none | [attack-L-law.md](history/staging/attack-L-law.md) |\n| `Q-L-subadditivity` | ANSWERED | Is L sub-additive when two prime sets merge, as the intuition that L should be smaller for the combined set suggests? | Read literally the intuition is wrong by a one-line proof rather than an edge case: L is super-additive, exhaustively censused to nine primes; read as a statement about reusing primes it is right, and that reading is worth more than the refutation. | none | [attack-L-subadditivity.md](history/staging/attack-L-subadditivity.md) |\n| `Q-T2a-prediction` | PARTIAL | What are G2(43#) and G2(47#), predicted blind from the thirteen exact terms? | Registered blind: G2(43#) central 630, window 561 to 747, and G2(47#) central 720, window 653 to 868, with the method's calibration PASSING on the eight-term prefix; the 43# term has since been enumerated at 618, inside the window and below the central value, and 47# remains unmeasured, with the report's own largest risk being that c2' may be about to turn down. | none | [phase1-T2a-prediction.md](history/staging/phase1-T2a-prediction.md) |\n| `Q-W1a-applied` | ANSWERED | Were the unapplied findings for partition A, the paper suite and citation-heavy files, applied? | Thirty-three findings, 24 applied, 3 refused at a primary source (two of which would have written a falsehood into the corpus) and 6 handed back; the most consequential item was not on the list, since covering-dive's stated reason for dismissing Kalmynin-Konyagin was false by the corpus's own proof and the corrected reason had to be found rather than copied. | none | [phase1-W1a-applied.md](history/staging/phase1-W1a-applied.md) |\n| `Q-W3-literature` | ANSWERED | What do the four literature targets give: the Omega(kappa, L) quantifier, the Kalmynin-Konyagin transfer, Erdos-Ruzsa 1980, and the two-class upper bound? | The quantifier is over all pairs, confirmed verbatim in two independent sources, and every page and section citation in beta2-note checks out; the higher-consequence find is that covering-dive's stated reason for dismissing Kalmynin-Konyagin is false, since the proof only sees the size of Omega_p, and the two-class upper bound survives a second independent attempt as ABSENT. | none | [phase1-W3-literature.md](history/staging/phase1-W3-literature.md) |\n| `Q-X-upper-0829n` | PARTIAL | Can the rough-pair census X(K) on the stretch window be bounded above, unconditionally and with an explicit error term, by a dimension-2 upper-bound sieve, and how loose is the bound against exact X(K)? | Yes, as a theorem with nothing open (Selberg's Λ² inequality with a remainder bounded by the interval structure alone, VERIFIED at all 10908 anchor-depth pairs), but it is loose by 7.82 to 10.13 against exact X(K) at the band depths, decomposed as a sieve loss 1.78 to 2.53 times a structural loss 4.01 to 4.39 that no upper-bound sieve can remove, the window holds the level below z² (s_max 1.81 to 1.96) so the bound stops improving past p_K = 31 to 61, its main term is the sifting function's and not the census's, and the drift is a crossing quantity outside the legal half and is not evaluated. | Z2 | [attack-0829n-X-upper.md](history/staging/attack-0829n-X-upper.md) |\n| `Q-a144311-vocabulary` | ANSWERED | In the vocabulary the literature actually uses, is there a published upper bound for G2, and is beta2 still the best known kappa = 2 sifting limit? | The terms of art are the paired Jacobsthal function and Jacobsthal-type functions for polynomials, and in that vocabulary no published upper bound exists at any exponent, beta2 is unimproved since 2008, nothing is found on A144311's asymptotic growth, and an Erdős-Rankin lower bound is free; both load-bearing corpus claims survive. | none | [audit-a144311-vocabulary.md](history/staging/audit-a144311-vocabulary.md) |\n| `Q-a3-05-bound-L` | PARTIAL | Can L be bounded unconditionally from run consistency, and where does that stop? | Theorem B gives the first unconditional bound, L <= 1 + m*, sitting 0 to 3 above the truth at all eight measured folds, but it is of order G2(old)/p, which is linear in x on the fold diagonal; section 8 locates the wall at one large-deviation hypothesis H″ the residue side cannot reach, and Theorem C inherits it for kappa(m). | none | [a3-05-bound-L.md](a3-05-bound-L.md) |\n| `Q-a9-histogram-operator` | ANSWERED | What is the exact evolution of the gap-size histogram under folding, as a transfer operator on count(d), and what is the tail P(gap >= 2p)? | The operator is derived and then verified by exact computation at every fold from T5 to T31 and the tail law is computed exactly, but the machinery is a rediscovery of Holt and Rudd arXiv:1408.6002 section 5 and may not be presented as new; only the G2 spacing beyond their \\|s\\| < 2p restriction is ours. | A9 (retired) | [a3-09-histogram-operator.md](a3-09-histogram-operator.md) |\n| `Q-ab-coupling` | MIXED (attack-ab-coupling.md: PARTIAL; verify-ab-coupling.md: ANSWERED) | How much of the counting criterion's known slack does the rigidity of the two kill channels A = 0 mod p and B = -2 mod p close? | The pilot survives, every held number reproducing on an independent code path, and the one correction is in its favour: the depth-3 ceiling is 38 slots, not 39; the LP answers NO, its exact depth-2 optimum over the full polytope being 54 slots, identical to the pilot's, so it prices the idea and does not clear 51. | none | [attack-ab-coupling.md](history/staging/attack-ab-coupling.md), [verify-ab-coupling.md](history/staging/verify-ab-coupling.md) |\n| `Q-adversary-wave2` | ANSWERED | Do the four same-evening headlines (kill shadow, scanstat, sofic, Shearer) survive a refute-first adversarial pass? | Nothing BROKEN: sofic SURVIVES and the other three SURVIVE WITH CORRECTIONS, the corrections being one verdict label a grade too high by the record's own prereg rule, three false custody sentences and an oversold prediction, a misdescribed extremal witness, and an atomicity hypothesis that is load-bearing and not shown. | none | [adversary-wave2.md](history/staging/adversary-wave2.md) |\n| `Q-advmin-1113` | CLOSED | Can joint caps quantified over classes recover the staircase-to-truth gap at @11 and @13? | CLOSED at @11: the adversarial minimum for the whole family is 16, below the class-blind ceiling and 18 under the staircase floor of 34, which also refutes the task's own framing that it would sit between floor and truth; at @13 the same family is priced at advmin <= 152. | none | [attack-advmin-1113.md](history/staging/attack-advmin-1113.md) |\n| `Q-agent-start` | ANSWERED | What must an incoming agent read and preserve before continuing the reviewed arithmetic campaign? | Onboarding and document ownership only. The second dispatch is completed and independently reviewed on 2026-09-09. RESEARCH-EXECUTION.md owns current dispositions and candidate next obligations; RESEARCH-HANDOFF.md owns the updated exact contract. No assignment is started by that review. Twin-prime infinitude and required signed estimates remain OPEN. | C | [AGENT-START.md](AGENT-START.md) |\n| `Q-anchored-calm` | PARTIAL | What is proven about the anchored calm, and what is not? | The calm names a phenomenon and never a claim: the Mirror-Sibling, Fusion, Mirror-Phase Doubling and Minus-Half results are PROVEN at all x and all scour q, the phenomenon itself is MEASURED (the anchored rotation at rank 2 of 510,510 at @17), and two things stay unproven, of which the Skeleton Equidistribution Conjecture is one. | none | [anchored-calm.md](anchored-calm.md) |\n| `Q-anchored-ladder` | ANSWERED | Does the unified anchored ladder generalize past @11 and @13, and what does cheap K buy at each level? | It generalizes verbatim to @17 and @19, and the K = 0 unified floor is DEAD there (vacuous at -1262 and -55865); what survives and grows is the family lead at the classic certification depth K*, margin +2, +5, +26, +110 across @11 to @19. | A | [attack-anchored-02.md](history/staging/attack-anchored-02.md) |\n| `Q-anchored-ladder-17-0830` | ANSWERED | At @17 and above, at which depth K does the composed (unified) anchored floor first turn positive and at which K does it reach the truth, and does the depth-cost curve K_first_positive(level), K_truth(level) move with level? | At @17/@19 the answer was already on record (Q-anchored-ladder: positive at K = 2/10, truth at ascending K = 109/410, minimal pool 88/350) and is re-derived here as the custody gate; the new level @23 ran under a sealed 14-of-14 forecast: positive first at K = 27 = K* (PROVEN from the record before the run), truth at ascending K = 1732 of 1739, minimal pool 1543 = 0.887 of the scour (PROVEN lower bound meets VERIFIED upper bound); across five levels positivity costs exactly K* while the truth fraction rises 0.40, 0.62, 0.73, 0.81, 0.89, a MEASURED finite table with no law in x and no exponent touched. | A | [attack-0830-anchored-ladder-17.md](history/staging/attack-0830-anchored-ladder-17.md) |\n| `Q-anchored-unify` | ANSWERED | Can the three anchored mechanics be unified at @11, and what floor family does the unified cap certify? | MEASURED and PROVEN-here but HELD: extending the staircase's cofactor injection to m = 1 folds the self-strike allowance into the count, the unified-cap lemma gives fresh <= capU_K <= cap_K at every depth and prime at @11 and @13, and the floor family runs 36 to 45; the x = 37 secondary is a clean negative. | A | [attack-anchored-01.md](history/staging/attack-anchored-01.md) |\n| `Q-anchored-vs-global-gap` | CLOSED | Is the anchored gap A(v) = F(v) - v easier to bound than the global gap G2(v#)? | The separation is real and correctly measured but worth exactly one unit of exponent against a deficit of 2.2665, and the anchored quantity's only proven bound is beta_2 = 4.26645, the IDENTICAL theorem rather than a consequence, because the DHR sieve is position-uniform; so the anchored object sits further from its bound, and Proposition D is a closure at today's exponent only. | none | [attack-block-09-anchored.md](history/staging/attack-block-09-anchored.md) |\n| `Q-anchored-windows` | SUPERSEDED | How much does anchoring at the origin enrich a window against a random window, and for how long? | SPENT, live treatment is paper/anchored-note.md: Theorems A-C proven (A and B unconditional, C's constants HL-conditional), anchored enrichment bounded by e^{2gamma} = 3.172 and mortal, the head-window dip constant e^{2gamma}/4 = 0.79305 with 0.89 a finite-size blend; the kill shadow is no longer a separate object and section 3's peak sequence remains unreproduced. | none | [anchored-windows.md](anchored-windows.md) |\n| `Q-applied-0828-backfill-1` | ANSWERED | Which legacy notes in chunk 1 received ledger blocks? | All 86 notes in the chunk received a block and none was skipped; 85 distinct question ids were minted, no id was reused from QUESTIONS.md section 2, and two notes were chained onto one shared id. | none | [applied-0828-backfill-1.md](history/staging/applied-0828-backfill-1.md) |\n| `Q-applied-0828-backfill-2` | ANSWERED | Which legacy notes in chunk 2 received ledger blocks? | 85 of the 86 notes in chunk 2 carry a block; only GLOSSARY.md was left unindexed, as a vocabulary registry that poses no question, and one id (Q-records-placement) was reused from QUESTIONS.md section 2 with its question text copied exactly. | none | [applied-0828-backfill-2.md](history/staging/applied-0828-backfill-2.md) |\n| `Q-applied-0828-backfill-3` | ANSWERED | Which legacy notes in chunk 3 received ledger blocks? | All 86 notes in the chunk carry a block; none was skipped, one id (Q-quadpoint-transplant) is reused from QUESTIONS.md section 2 with its question text copied exactly, one id (Q-f-census) is shared by the two notes of the f-from-census chain, and seven blocks name a TODO item whose Ledger line does not yet list them. | none | [applied-0828-backfill-3.md](history/staging/applied-0828-backfill-3.md) |\n| `Q-applied-0828-backfill-4` | ANSWERED | Which legacy notes in chunk 4 received ledger blocks? | All 85 notes of chunk 4 carry a ledger block, none reuses an id already in QUESTIONS.md section 2, and none was left unclassified. | none | [applied-0828-backfill-4.md](history/staging/applied-0828-backfill-4.md) |\n| `Q-applied-0828-census` | ANSWERED | Were the census red team's corrections applied to the three HELD notes? | Every WEAKENED and REFUTED verdict in redteam-0828-census.md is applied to its target note in the present tense; two producer defects and four live-doc corrections stay open and are listed here. | Z0, Z4, Z6 (retired) | [applied-0828-census.md](history/staging/applied-0828-census.md) |\n| `Q-applied-0828-closures` | ANSWERED | Were the closures red team's corrections applied to the five HELD notes? | Applied. 47 prose corrections across the five notes; c2prime-refit-22 drops from CLOSED to PARTIAL and three ledger verdicts are rewritten; four producer READINGS blocks and eleven live-document lines carry the same defects and are listed here uncorrected, behind the fence. | 10 (retired), 1c (retired), 1d, Z5b (retired), 0 | [applied-0828-closures.md](history/staging/applied-0828-closures.md) |\n| `Q-applied-0828-engine` | ANSWERED | Were the engine red team's corrections applied to the five HELD notes? | Applied, 37 edits across the five notes: two numbers corrected (first s >= 10.82 at x = 263 not 239, first non-empty kappa=1 level @37 not @53), two ledger verdict lines and one note title rewritten, one proof step given BV's max-over-y form, the one-class freshness factor restated on P^-(m) >= q so it holds at every scour prime, and two mislabelled columns renamed; six corrections that land on live documents are collected here unapplied, and no producer was touched. | 11 (retired), 8 | [applied-0828-engine.md](history/staging/applied-0828-engine.md) |\n| `Q-applied-0828-glossary` | ANSWERED | Was the hyperuniformity conflation corrected? | In the vocabulary yes: GLOSSARY's single entry is split into Sub-Poisson window counts (Var/E), which keeps every measured value, and Hyperuniformity, which now carries Torquato's definition, the class I/II/III taxonomy, the tile's vacuous class I by periodicity and the sub-Poisson-not-hyperuniform reading at L = y^u; in the corpus no, one staging note is corrected here and ten further loose uses in nine .md files plus eighteen in seven scripts sit outside the fence with their replacement sentences listed below. | none | [applied-0828-glossary.md](history/staging/applied-0828-glossary.md) |\n| `Q-applied-0828-head` | ANSWERED | Were the head red team's corrections applied to the three HELD notes? | Applied. Every WEAKENED and REFUTED row of redteam-0828-head.md lands in the note it targets as 42 prose edits, and all three ledger verdict lines now carry the surviving chain; four corrections fall outside the fence and are owed elsewhere, the load-bearing one being destroyer-census-01 §6(b)'s R and its h/R = 1.09 -> 1.03. | Z4 | [applied-0828-head.md](history/staging/applied-0828-head.md) |\n| `Q-applied-0828-litimports` | ANSWERED | Were the lit/imports red team's corrections applied to the six HELD notes? | Applied, 64 substantive edits across the six notes: the Saffari-Vaughan range exponent is 6/11 in both import notes, the fixed-M THEOREM in import-fracparts is struck and the cap-36 MEASURED to PROVEN upgrade withdrawn, row 15 loses its theorem column and row 16 is regraded stale, and the six weakened sentences in the four literature notes now read as the audit requires; four claims are declined with reasons, and nine live documents plus two fenced staging notes are recorded here as owing corrections that this pass did not apply. | Z5, 4 (retired) | [applied-0828-litimports.md](history/staging/applied-0828-litimports.md) |\n| `Q-applied-0828-live` | ANSWERED | Which engine and head red-team corrections reached the live layer? | Nineteen edits across four documents: the glossary's cap_K dimension count with its mandatory sifted-versus-assumed clause plus three new entries (twin opener, head, tail); the survey's one-class density factor, S1's restored pi(q-1) terms and r in N_x hypothesis, the dropped Chen absorption, Theorem C with its emptiness, dimension 1 for q_i < q, the comb-conditioning qualifier and the inverted shallow-band bullet; the certificate engine's equidistribution status cell, the q = T^{o(1)} second hypothesis and the Comb Discrepancy scope cell; and the census's R as the continuum functional with its three comparators; paper/staircase-note.md, TODO.md and QUESTIONS.md are owed and untouched. | 11 (retired), 8, Z4 | [applied-0828-live.md](history/staging/applied-0828-live.md) |\n| `Q-applied-0828-quadpoint` | ANSWERED | Were the quadpoint red team's corrections applied to the four HELD notes? | Applied: the nine REFUTED and WEAKENED verdicts that target the four notes are written into them in present tense, with ledger blocks added; the tenth targets the sealed prereg, outside the fence, and is recorded in attack-quadpoint-02.md instead; three corrections owed by TODO.md Z2 and one by Z0 are listed here and are NOT applied. | Z0, Z2 | [applied-0828-quadpoint.md](history/staging/applied-0828-quadpoint.md) |\n| `Q-applied-0828-queue0827` | ANSWERED | Were the 2026-08-27 drafted corrections applied? | Eighteen edit entries applied across six files, the Wu band-label inference and the saturation claim being the load-bearing ones; six proposals declined, two on ownership, one because the brief forbids promoting its note, one as already carried in paraphrase, one on an unverified inferred theta-law, and one because two scratchpad notes disagree on the number. | none | [applied-0828-queue0827.md](history/staging/applied-0828-queue0827.md) |\n| `Q-applied-0828-registries` | ANSWERED | Which lit/imports corrections reached the registries? | 13 edits across six files, plus one appended changelog section: IMPORT-MAP gains rows 15 (no THEOREM, the column struck), 16 (STALE) and 17 (CLOSURE, two-sided) with its three counts recounted to seventeen rows; SEARCH-CONVENTIONS gains the Kourbatov-b, generalized-Dickman and VC-of-multiples rows with each channel's calibration state; import-map-construction's determinantal cell, natal-cap-36's separate-census sentence, one REFUTED row and one PRIOR-ART block are corrected; GLOSSARY, TODO, QUESTIONS, the five loose hyperuniformity documents and two staging notes are left for their holders and listed here. | Z5, 4 (retired) | [applied-0828-registries.md](history/staging/applied-0828-registries.md) |\n| `Q-applied-0828-registries2` | ANSWERED | Which queued registry corrections were applied? | 15 edits across four files: SEARCH-CONVENTIONS gains four §1 rows (the census as a pretentious correlation with the Chowla warning, Boolean analysis on the CRT product, covering a cyclic group by translates, the fractional parts of N/n) and five §3 rows, plus one pipe fix that was hiding a clearing phrase from the gate; import-rough-anatomy's λ₁ → λ₂ comparand is corrected in four places, 12–23× becoming 2.11 to 2.86×; lit-dickman-variance §6.3's \"row 15\" is disambiguated to the drafted row it means; IMPORT-MAP's circularity and payoff tallies are recounted and now sum to seventeen; REFUTED.md needed no edit and got none. | none | [applied-0828-registries2.md](history/staging/applied-0828-registries2.md) |\n| `Q-applied-0828-todo-prune` | ANSWERED | Which TODO items left the file under its charter after 2026-08-28? | Five left (11, 10, 1c, 1e, Z5b), nine were restated on what remains (Z2, Z4, Z5, Z6, Z7, Z0, A, 0, 1d, 4, 8), the rest stand; every retained Ledger line is verbatim, no claim's rung moved, and eleven notes now need `(retired)` on their todo lines before the gate is green. | none | [applied-0828-todo-prune.md](history/staging/applied-0828-todo-prune.md) |\n| `Q-applied-0828-varE` | ANSWERED | Were the Var/E red team's corrections applied to the four HELD notes? | Every WEAKENED and REFUTED verdict in redteam-0828-varE.md is applied to its target note in the present tense; four ledger verdicts are rewritten, the theta=1 row is raised to PROVEN, and seven live-doc sentences plus one producer reading stay open and are drafted here. | 9 | [applied-0828-varE.md](history/staging/applied-0828-varE.md) |\n| `Q-applied-0829-ledger` | ANSWERED | Were today's notes linked into the question ledger's TODO guard, the Z2 draft's last two non-fit lines fixed, and the covering-prune root corrected? | All three landed and none of them moves a result: nine of thirty notes read now name the TODO item their move executes and four TODO items carry the new ids, so `node research/qc.js` reports the ledger check clean at 0 findings; the Z2 draft's legal items 3 and 4 carry the audit's instrument rider and the quantifier-police clause the red team asked for; and the covering-prune producer now computes the root of a*ln(a/e) = 1 instead of quoting it, giving a = 3.5911214767 and a divergence constant of 7.1822 lnln x against the 7.19 the file carried, a 0.08 per cent change with nothing downstream consuming it. | none | [applied-0829-ledger.md](history/staging/applied-0829-ledger.md) |\n| `Q-applied-0829-measure-a` | ANSWERED | Were red team A's corrections applied to the notes it reviewed? | Of the nine APPLY items in redteam-0829-measure-a.md §3a-3c, seven were applied here as exact string edits, two (E1 and E5) were verified as already applied by the producing agent with a producer fix and a forced re-embed, one clause of E1's proposed wording is left unapplied and is recorded as owed, and the four live-document items of §3d plus the QUESTIONS.md regeneration are left for the orchestrator; node research/qc.js returns TOTAL 0 across all thirteen checks after the edits. | none | [applied-0829-measure-a.md](history/staging/applied-0829-measure-a.md) |\n| `Q-applied-0829-measure-b` | ANSWERED | Were red team B's corrections applied to the notes it reviewed? | 17 of 17 APPLY items whose target is a staging note are applied in place, 2 ledger verdict lines rewritten as the red team specified (measure-record-null2-0829 and zone-tail-02-0829), 3 HOLD/DECLINE items deliberately not applied, 5 live-document items left verbatim for the orchestrator, and 3 producer items recorded as OWED without touching any .js; the fast gate returns 0 findings after the pass, the single mid-pass finding having belonged to another agent's file, research/measure-g2z2-0829.js:432, and never to any edit made here. | none | [applied-0829-measure-b.md](history/staging/applied-0829-measure-b.md) |\n| `Q-applied-0829-measure-c` | ANSWERED | Were red team C's corrections applied to the notes it reviewed? | 22 of the 25 items in section 3 target the four HELD staging notes and all 22 are applied (21 as the exact old-text to new-text strings the red team wrote, 1 as a minimal carry because filling its column for all nineteen rows would be a fresh grading judgement rather than a citation), plus 1 directed edit to lit-scourfield-2008.md section 0 that red team section 1g names and section 3 gives no string for; 3 items are flagged HOLD and are left verbatim for the orchestrator (varE-theta2-proof.md, lit-smooth-divisors.md, SEARCH-CONVENTIONS.md); 1 producer item is OWED and untouched, since the producer's readings 13 and 14 still carry the M-P1 failure that E8 and E9 reverse in the note and correcting them needs a re-embed; node research/qc.js returned TOTAL 1 on the run straight after the edits, the one finding being research/measure-g2z2-0829.js:432 in another agent's file, and TOTAL 0 on the run after this note was written, with refs, quotes and ledger clean on both. | none | [applied-0829-measure-c.md](history/staging/applied-0829-measure-c.md) |\n| `Q-applied-0829-refuted` | ANSWERED | Were the four refuted-claims audits' APPLY row rewrites applied to REFUTED.md? | 25 of the 27 APPLY items are in the file, 1 more merged into a sibling's text at the shared boundary row, 1 deferred to the in-progress red team, 7 HOLD items untouched and listed here for Chris, 4 items owed outside REFUTED.md; the table still holds exactly 67 data rows and qc.js reports refs 0 and crosslinks 0. | none | [applied-0829-refuted.md](history/staging/applied-0829-refuted.md) |\n| `Q-applied-0829-registries` | ANSWERED | Were the registry debts of 2026-08-29 (IMPORT-MAP rows, the PRIOR-ART spectrum line, the escape labels, the REFUTED HOLD proposals) landed? | All four debts are discharged as far as a non-adjudicating pass may take them: IMPORT-MAP carries rows 20 and 21 at CLOSURE-plus-anchor and anchor-plus-wall-address with nothing opened, PRIOR-ART's Holt-Rudd attribution now names the printed spectrum and the SCRATCHPAD-GRADE 2.82/ln p mixing rate with priority stated as Holt and Rudd's, recon-0829-escapes.md section 6 carries all sixteen owed (i)/(ii)/(iii) labels of which nine are fresh judgements and three come out (ii), and three of the seven REFUTED HOLD proposals were applied as wording and calibration while four are left verbatim for Chris because they touch a producer, a canonical file, a cross-file number or the index's format; the table stands at 67 rows and qc.js is 0 on refs, crosslinks, calibration and search-convention. | none | [applied-0829-registries.md](history/staging/applied-0829-registries.md) |\n| `Q-applied-C` | ANSWERED | Which wave-2 consistency findings were applied to the paper suite? | Applied across the seven paper/ files with post-edit line numbers, plus a CHRIS-DECIDES list, handoffs to other partitions and one unresolved item: two scopes for the Cov_adj anticorrelation exception count, both now printed, one of which should be checked and the other dropped. | none | [applied-C.md](history/staging/applied-C.md) |\n| `Q-applied-G` | ANSWERED | Which wave-3 partition G findings were applied, and what did the pass find on its own? | Applied to the eleven owned files: the OEIS off-by-one confirmed and provable four ways, a factor of two found in the seam draft's Hardy-Littlewood heuristic, the L gap corrected in three files from a flat 0.58 ln p to the home's 0.58 to 0.95 ln p range, seven orphans swept, and one supplied answer refused as wrong before it reached a header banner. | none | [applied-G.md](history/staging/applied-G.md) |\n| `Q-applied-Q` | ANSWERED | Do the two OEIS drafts survive an instrument that recomputes them rather than reads them? | Two defects found in the G2 draft (the Maier-Pomerance conjecture cited in the wrong direction, and a no-upper-bound-published claim the repository itself contradicts) plus a third error class routed from partition P; every DATA term was recomputed and audit-numbers went from 37/39 to 76/78 checks. | none | [applied-Q.md](history/staging/applied-Q.md) |\n| `Q-applied-wave2-A` | ANSWERED | Which wave-2 partition A corrections were applied, and to which files? | Five jobs applied across G2-STATE.md, LOCALIZED-GAP.md, maier-matrix.md and origin-excess.md, with partition A's own qc columns at 0 on every check afterwards; handoffs to other partitions and one decision needing Chris are listed rather than guessed. | none | [applied-A.md](history/staging/applied-A.md) |\n| `Q-applied-wave2-B` | ANSWERED | Which wave-2 partition B corrections were applied to U-FRAME.md, ATTACKS3.md and gate-multiplies.md? | Applied with the U-FRAME section map preserved (128 inbound pointers, 47 on 5a, nothing renumbered), the strategic contradiction resolved, statistics corrected, one dead quotation removed and two further corrections to the record found on the way; section 12 lists what stayed unresolved. | none | [applied-B.md](history/staging/applied-B.md) |\n| `Q-applied-wave2-D` | ANSWERED | Were wave-2 partition D's findings applied to the eight entry-path documents? | Applied across all eight files with qc total 22 to 21 and one refs finding of its own (a framework bug), two transfers pairs word-diffed; the em-dash scope question and one GLOSSARY entry are flagged as judgement calls rather than omissions, and other partitions were live in the same tree so the deltas are not all this partition's. | none | [applied-D.md](history/staging/applied-D.md) |\n| `Q-applied-wave2-E` | ANSWERED | Which wave-2 partition E corrections were applied, and to which files? | Seven jobs applied across covering-dive, two-class-lower-bounds, PRIOR-ART, sift-limit-attack, theta-ladder, maxgap-law and dhr-verification, with refs and quotes at 0 in this partition before and after; one novelty verdict in a publication-track file is weakened by de Polignac 1849 and flagged for Chris. | none | [applied-E.md](history/staging/applied-E.md) |\n| `Q-applied-wave2-F` | ANSWERED | Which wave-2 partition F corrections were applied to the natal-cap family and its compound claims? | Applied across nine natal-cap files with two new homes created (anchored-calm.md and certificate-engine.md); the Fused-Window Calm Lemma was decomposed and its name retired, the X-limitation theorem's scope is confirmed a scope error, and the refutation is recorded as bigger than it had been reported. | none | [applied-F.md](history/staging/applied-F.md) |\n| `Q-applied-wave3-H` | ANSWERED | Were wave-3 partition H's U-FRAME sections 10-12 split into child documents? | Split and applied: U-FRAME falls 1,032 to 772 lines with three children created, but the corpus as a whole is 87 lines larger and that must not be reported as a shrink; refs and crosslinks stay at zero and no transfers finding exists between any stub and its child. | none | [applied-H.md](history/staging/applied-H.md) |\n| `Q-applied-wave3-I` | ANSWERED | Was the flagship paper split as wave 3 partition I directed, and what did the split expose? | Split into paper/moire-primes.md and the new paper/wall-note.md, with every deliberate change to moved text word-diffed; three open defects sit inside the split, two of them already landed and one not, and the judgement calls are listed so they can be overruled. | none | [applied-I.md](history/staging/applied-I.md) |\n| `Q-applied-wave4-P` | ANSWERED | Which calibration defects sit on a3-05-bound-L.md, localized-04-maxsum.md and level-ledger-tight.md, and were they fixed? | Fourteen defects found and fixed on the one axis wave 3 never ran, none of them reachable by any check in qc.js: two numbers contradicting the file's own pasted script output, one slot count against thirteen other sites and an exact closed form, one superseded sifting limit and three dropped scopes; the gate stayed at zero on all six checks. | none | [applied-P.md](history/staging/applied-P.md) |\n| `Q-at43-bigint-0830` | PARTIAL | Can the K-30 natal march be carried past its 2^53 ceiling to @43 exactly, and is the @43 point worth its cost? | The engine side is done and VERIFIED (five paths promoted to BigInt, @7..@37 and a 0.248% slice of @41 reproduced digit for digit); the @43 point is NOT run, because the measured extrapolation is 579 h of eight cores on this machine (a 43x tile times a 2.23x per-cell cost, against the brief's ~13x), which is a weeks-class box job whose only payoff is a sixth point on a curve with no consumer; the forecast 0.8393 raw / 0.8399 persisted stands unscored. | 1 | [engine-0830-at43-bigint.md](history/staging/engine-0830-at43-bigint.md) |\n| `Q-attacks-wall-ten` | ANSWERED | Which ten attacks on the frontier-zone wall were executed, and what did each return? | All ten are marked DONE with their learnings recorded; none breached the wall, parity stands, and the durable yields are attack 8's unconditional small-gap theorem for p >= 17 and the anchored-origin phase picture. | none | [ATTACKS.md](ATTACKS.md) |\n| `Q-attacks2` | ANSWERED | What did the ten copy-geography attacks of the second wave find? | All ten complete: one theorem-grade result (the Exact Invariance Lemma, fossil depth frozen at birth, proved independently twice) and one MEASURED, Hardy-Littlewood-conditional law (the Unification Law, to about 1 percent); no unexplained novelty survives and the wall stands. | none | [ATTACKS2.md](ATTACKS2.md) |\n| `Q-attacks3-uframe` | ANSWERED | What did the ten-attack u-frame wave (A1-A10) settle about the fold structure and the Zone Postulate? | The R/I split held: every residue attack returned an exact object and no interval attack moved; the additive chain G2(new) <= G2(old) + L*mbar it was framed around is itself REFUTED (U-FRAME 5a step 4), A5's Theorem B is structurally capped, and A10's exactness route is a dead end (rate 0.927 over 329 cells, excess 6..120 not shrinking). | none | [ATTACKS3.md](ATTACKS3.md) |\n| `Q-audit-campaign` | ANSWERED | Does the campaign archive layer - the natal-cap notes, ATTACKS, covering-dive and their neighbours - hold internal inconsistencies? | Six files and eight of the natal-cap notes are clean and need no change; the remainder carry per-file findings recorded in place, and this is a diagnosis pass rather than an edit. | none | [audit-campaign.md](history/staging/audit-campaign.md) |\n| `Q-audit-cross-doc` | ANSWERED | Where does the corpus carry two values for the same object in two different documents? | Twenty-six findings ranked A, B and C, eighteen of them the Kourbatov shape where the resolution was already on disk in a document the stale one cites by name, including a retracted theta column quoted as live in nine places and a prior-art collision on the dimension-2 Rankin accounting; the gate read 0 findings before and after. | none | [audit-cross-document-constants.md](history/staging/audit-cross-document-constants.md) |\n| `Q-audit-front-door` | ANSWERED | Does someone arriving after the 2026-08-19/20 wave get the true current picture from README, G2-STATE section 0 and CHRONICLE in ten minutes? | All three are STALE: README's Status is dated 2026-08-17 with a broken map row and four missing files, G2-STATE section 0 stops at the second attack wave, and CHRONICLE ends at Day 5; drafts are ready for two of the three and README's rewrite is left in Chris's register. | none | [audit-front-door.md](history/staging/audit-front-door.md) |\n| `Q-audit-new1` | ANSWERED | Do G2-STATE.md, LOCALIZED-GAP.md, localized-04-maxsum.md, exponent-control.md and gate-multiplies.md agree with themselves and with their sources? | Findings per file, the leading one a NARROWED Origin Excess claim: the lemma carries an unstated third hypothesis, positivity of D_y(y'^2) - 2(pi(x) - pi(y) + 1), which forces ln x*/ln y near 1.44 at every level computed and caps the advantage at about 2.2 rather than order (ln x / ln y)^2. | none | [audit-new1.md](history/staging/audit-new1.md) |\n| `Q-audit-new2` | ANSWERED | Are the seven research documents of the new2 partition internally consistent and correctly scoped? | Per-file findings across the seven, mostly dropped scopes and attribution defects, plus four documents that were missing the standard current-understanding footer, which was added. | none | [audit-new2.md](history/staging/audit-new2.md) |\n| `Q-audit-numbers` | ANSWERED | Does the corpus's arithmetic survive recomputation from scratch by an independent method? | The ladder is right: all twelve G2 terms reproduce under a method sharing no code with the one that produced them, 37 of 39 assertions pass in 152 s, and the four errors found are small and none is load-bearing. | none | [audit-numbers.md](history/staging/audit-numbers.md) |\n| `Q-audit-papers` | ANSWERED | Are the paper suite and the root README consistent with their research homes? | Per-file corrections applied across the six paper/ files and README, with a Not fixed section naming what was deliberately left, including the exposition question the Holt attribution raises and one stylistic sentence of Chris's that was left alone. | none | [audit-papers.md](history/staging/audit-papers.md) |\n| `Q-audit-reading-path` | ANSWERED | Can a cold reader reach every live result in two hops? | The two-hop test passes for eight of the ten headline results of 2026-08-19/20 and no staging record is orphaned; what fails is narrower, the front door being three days behind and contradicting REFUTED.md on u_sup, G2-STATE's section 10 table still the 2026-08-17 list, and kappa-not-L carrying nothing from the L wave. | none | [audit-reading-path.md](history/staging/audit-reading-path.md) |\n| `Q-audit-spine` | ANSWERED | Do the spine documents - ZONE-POSTULATE, THE-DIALS, THE-LENS, GLOSSARY, OBSERVATIONS - state their objects correctly? | Findings per file, the leading one a frame error in ZONE-POSTULATE section 5: G2's own exponent measures 1.57 central, bracket 1.3 to 1.9, against a calibrated one-class control, while theta is above 2 and rising, so the two readings that sat on boundary exponent 2 resolve in opposite directions. | none | [audit-spine.md](history/staging/audit-spine.md) |\n| `Q-audit-staleness` | ANSWERED | How much of the live layer is stale against changelog passes 26 to 48? | Eleven of the nineteen files in scope are clean and eight carry at least one stale sentence (U-FRAME three, G2-STATE two and others), with a separate list of upgrades the live layer has not absorbed; report only, nothing outside the report was edited. | none | [audit-staleness.md](history/staging/audit-staleness.md) |\n| `Q-audit-uframe` | ANSWERED | Do the u-frame documents' claims survive an audit against their own sources? | Several are REFUTED: the multiplier requirement ln c <~ 2 ln^2 u / u is off by a factor ln u and is replaced by the per-fold rate ln c(p) <= 2 ln p / p, the Inertness Lemma's rate is narrowed to 4/u, and SUM L*mbar ~ 2 x ln^3 x is false because U-FRAME step 4 is false. | none | [audit-uframe.md](history/staging/audit-uframe.md) |\n| `Q-b2mean` | ANSWERED | Can the mean value of B2 = Sum e^2 Vabs(e)^2 be bounded below its trivial qmax*B ceiling? | Yes, by one logarithm: B2 = O(qmax*ln^7 z) against the trivial O(qmax*ln^8 z), with the mass located on primorial-core smooth e nowhere near qmax, the primitive-mode weight J2(e)/12 proved and priced as a flat 1.48x cut of MS3, and the finite forms computed exactly at z = 13..47. | none | [attack-b2mean-01.md](history/staging/attack-b2mean-01.md) |\n| `Q-barrier-kappa2` | CLOSED | Does Granville's exceptional-character construction, which makes the Jurkat-Richert f and F attained by interval sieving problems at kappa = 1, transfer to kappa = 2 and supply the missing barrier on wall-note Face 4, so that no class-blind argument can beat beta_2 = 4.26645 on intervals? | No. Verdict (b): the transfer runs, fixes the axiom obstruction that sift-limit-attack.md section 2 blamed for the mixed-sign sketch's death, and then hits a hard ceiling at u = 2, because the character's only gift is the one-point identity lambda = chi on z-rough integers and a sign pattern with a +1 coordinate is empty exactly when z^2 >= x; above u = 2 the surviving question is the shifted intersection of two positive-density lambda-classes, which is Chowla-strength, so the section 2 verdict stands with a corrected reason; the band (2, 4.26645] is untouched, and the hypothesis itself implies the target by Heath-Brown 1983. | 0 | [attack-barrier-kappa2.md](history/staging/attack-barrier-kappa2.md) |\n| `Q-bc-parity-floor` | CLOSED | Is the 2 that the census's B/C ratio limits to the same object as the parity floor's 2? | REFUTED at the first step, before any kappa-scan: the certificate dies where B/C crosses 1, not 2 (B = 503 against C = 497 at the death zone), the constant reads the sifting dimension while the threshold does not, and the parity floor's 2 cannot be this one. | none | [attack-bc-parity-floor.md](history/staging/attack-bc-parity-floor.md) |\n| `Q-beta2-AB-bounded` | ANSWERED | Is B(z,s) bounded, and does bounding it move the 4.2665 exponent? | B(z,s) <= 9 A(z)^2 (E(z) - 1) = O((log z)^8) is proven, unconditional, explicit and uniform in s, but the gap it closed was never load-bearing, the estimate has a published neighbour at s >= 9, Opera de Cribro Lemma 6.18, which does not cover this corpus's s in [2.0, 3.4], and the route to B = O(1) is measured dead as a uniform statement. | none | [attack-AB-bounded.md](history/staging/attack-AB-bounded.md) |\n| `Q-beta2-exact-strata` | CLOSED | Does re-inserting exact Buchstab strata (DP3) lower the dimension-2 positivity threshold below beta2 = 4.26645? | A sharp negative with a price tag: exact strata are legitimate and worth 1.72 to 1.84 of exponent, taking the positivity threshold to 2.0260 at finite levels, but they produce no theorem because the exact stratum value is the upper sieve function of a dilated tile one level down, so the regress terminates at every finite z and at no uniform z. | none | [attack-beta2-03-exact-strata.md](history/staging/attack-beta2-03-exact-strata.md) |\n| `Q-beta2-loss-budget` | ANSWERED | Where in the chain from truth to beta2 = 4.26645 is the loss, and which discard point is worth attacking? | Aim at theta: it carries the largest derivative (-4.2665 per unit) and the only consumer in the literature, beta is worth at most 1.07 and has no consumer, DP2 and the o(1) are worth exactly zero, and the two live knobs together stop at 1.597, so the last factor of 1.6 down to truth is in neither of them. | none | [attack-beta2-04-loss-budget.md](history/staging/attack-beta2-04-loss-budget.md) |\n| `Q-beta2-theta-total` | CLOSED | What is the largest theta_total defensible position-uniformly, and does it clear the break-even the beta2 upper-bound attack needs? | The largest defensible theta_total is 1 and it does not clear break-even: it misses by 0.2417 under the brief's symmetric arithmetic and by 0.2090 under the corrected asymmetric one, no part of the gap closes, and the weakest link in the negative is an ABSENT rather than a PROVEN. | none | [attack-beta2-02-theta-total-minimal.md](history/staging/attack-beta2-02-theta-total-minimal.md) |\n| `Q-beyond-chebyshev` | PARTIAL | Can the ensemble bound on P(S = 0) be pushed beyond Chebyshev, and how far? | Two beyond-Chebyshev bounds now stand, at @11 where capacity gives P(S = 0) = 0 exactly and at @13 where P(S = 0) <= 1.898e-6, a factor 513 below Chebyshev's 9.74e-4, with 683 assertions machine-checked; @17 is blocked on a 4.9e16-term sum and needs a 4-point wrap identity that does not exist. | none | [natal-cap-21-beyond-chebyshev.md](natal-cap-21-beyond-chebyshev.md) |\n| `Q-bf-split` | ANSWERED | Is Brüdern-Fouvry's level split optimal for their own theorem, or is there unclaimed exponent in it? | It was 99.3 per cent a trap: their condition (iii) binds at the published split and the proposed re-split violates it by x^{0.040403}, leaving 0.0000120536 in theta of the claimed 0.0295482779, real but invisible at the four decimals they print. | none | [attack-bf-split.md](history/staging/attack-bf-split.md) |\n| `Q-bilinear-fold-first-attack` | ANSWERED | Does assigning each Mobius factor to its first fold, or applying a direct second-moment bound, establish the missing one-sided estimate? | The first-fold identity yields no saving; a direct second moment has a negligible diagonal and an open two-prime off-diagonal. Fixed-fold covariance is an exact residue-imbalance square and the beta weights balance at each fixed modulus. Growing-depth joint control remains OPEN; B>=-C2*x+o(x) is not improved. | C | [bilinear-fold-attack.md](bilinear-fold-attack.md) |\n| `Q-bilinear-transplant` | CLOSED | Does this programme's frame have anywhere to put bilinear or Type-II arithmetic information, the only thing that defeats parity in practice? | The tile side cannot host it, since no corpus object is a signed sum over two ranging element ranges; the zone side can express one, and the estimate that one needs is the one already recorded absent. Section 9 lists the four findings that would reopen the row. | none | [attack-bilinear-transplant.md](history/staging/attack-bilinear-transplant.md) |\n| `Q-block-L-first-dead` | ANSWERED | Is the block-1 counting bound on L the first-dead 62 or the last-alive 111, and does downward closure hold in the block coordinate? | Downward closure HOLDS in the block coordinate: L <= 62 is sound and recomputed from scratch rather than inherited, L <= 111 is true, implied by 62 and weaker by 1.79x, and the adjudication's supposedly sharper instrument is attack 3's instrument. | G (retired) | [adjudicate-L-bound.md](history/staging/adjudicate-L-bound.md) |\n| `Q-block-alternation` | CLOSED | Is there a block-level alternation law? | No: a block-level inequality exists and is a real theorem, but it is not an alternation law - over a block the alternation law is strictly weaker than plain slot counting by the exact factor mbar(T_v)/6, at least 10/6 and growing like ln^2 v, and the resulting Block Partition Inequality closes only at v = 5 and v = 7 and is vacuous from v = 11 because 2*Sum_{v v^2 block ladder to sustain itself? | No: on the eleven-term exact ladder (this attack computed G2(41#) = 546) the ratio is flat at slope +0.05 +/- 0.11 on b >= 11 against a 12.3 per cent band, while the same estimator resolves the one-class control's falling ratio at 11.6 sigma; and the criterion is not independent, since at the natal it IS G2(x#)/x^2, the Zone Postulate itself. | none | [attack-block-01-ladder.md](history/staging/attack-block-01-ladder.md) |\n| `Q-block-exponent` | CLOSED | Does the block sub-problem have a better exponent than the proven 4.2665? | The required exponent for the ladder to close is 2, not 4 (the brief's step-3 arithmetic carried a factor error), and 2 is the TPC-equivalent line, so the useful part of the open band (2, 4.2665] is empty; the block's own prime range pins kappa = 2 and starting from T_v does not change the dimension. | none | [attack-block-05-exponent.md](history/staging/attack-block-05-exponent.md) |\n| `Q-block-greedy` | ANSWERED | Does the block-restricted adversarial greedy reach the truth at v = 5, and does block restriction cost anything? | Deterministic greedy reaches 17 against truth 19, randomised restarts reach 19 and an exhaustive DP confirms it, so greedy as a rule is optimal at v = 5 and the shortfall is a search failure; block restriction costs nothing on ten points with the same estimator on both sides, and the published Y2 ladder's bisection on a non-monotone predicate is a real defect. | none | [attack-block-04-greedy.md](history/staging/attack-block-04-greedy.md) |\n| `Q-block-kappa` | CLOSED | What does the block version of kappa(m) buy over the per-fold form? | Nothing: kappa_block(1) = 19 equals combined L at block 1 by construction, the minimal shift j*(m) equals kappa_block(m) at every m from 1 to 12 so there is zero slack in the block object, and the block form of Theorem C is empty, its vacuity coefficient 1.552 against a requirement of 1 at block 1 and growing like ln^2 v. | none | [attack-block-07-kappa.md](history/staging/attack-block-07-kappa.md) |\n| `Q-block-killers` | ANSWERED | For a maximal run of deleted T_v slots under the block v -> v^2, which primes do the deleting and how concentrated is the work? | Not concentrated: across 40 runs in five datasets every block prime kills at least one slot of every run and the exact minimum set cover is 82 to 100 per cent of the block, so no bound ranging over the block gets a reduction; the one favourable number is 915 of 1146 slots, 79.8 per cent, with exactly one killer. | none | [attack-block-02-killers.md](history/staging/attack-block-02-killers.md) |\n| `Q-block-literature` | ANSWERED | Does the literature already own the range-restricted covering object behind the block v -> v^2? | FOUND, and it has a name: the range-restricted covering problem is Erdős Problem #688 and is OPEN, the bounded-capacity conjecture #1200 is a third trap of the same class as the brief's two, and the Kalmynin-Konyagin journal version that covering-dive.md says does not exist does; the remaining negatives are ABSENT-AFTER-COMPETENT-SEARCH on calibrated channels. | none | [attack-block-06-literature.md](history/staging/attack-block-06-literature.md) |\n| `Q-block-reformulation` | CLOSED | Does the v to v^2 block reformulation, treating the block as a set, give a smaller combined L and an exponentially growing window? | Sound reformulation, no open path: no primality input is needed anywhere, nothing is deleted inside the window at all so combined L there is identically zero, and the object has to live in the tile; this note's own correction (the counting bound is 111, not 62) was itself reversed the same day by adjudicate-L-bound.md. | none | [attack-block-00-ADJUDICATION.md](history/staging/attack-block-00-ADJUDICATION.md) |\n| `Q-block-secondmoment` | CLOSED | Can a second moment over a block bound the empty windows, and is it allowed to work? | NO, on a tile-independent impossibility: Chebyshev needs dist(E, Z) < E/sqrt(W) and the measured shortfall at H = 204 is a factor 8.41e6, so a second moment gives a density bound that is legal at every H and TPC-relevant nowhere; the TPC line sits at eps = 1/W and at H about w^2, and quantifying over w buys nothing. | none | [attack-block-08-secondmoment.md](history/staging/attack-block-08-secondmoment.md) |\n| `Q-block-target` | CLOSED | What exponent does the squaring ladder v -> v^2 actually need, and do the steps get easier up the ladder? | The required exponent is exactly 2 with a constant below 1, which by the p^2 rule is TPC-equivalent, and the requirement is exponent-invariant at every step, so as a route to a theorem weaker than TPC the ladder is dead. | none | [attack-block-10-target.md](history/staging/attack-block-10-target.md) |\n| `Q-bonferroni-degree` | CLOSED | What degree does the Boole-Frechet certificate need, and does raising the degree reach an asymptotic bound? | Dead for asymptotics on a loop that closes on itself: raising the degree divides the over-certification exponent by the number of pairs without removing it (measured slopes 0.5441, 0.5425, 0.2816, 0.2937, 0.1695 at k = 2..6 against the predicted 1/(2 floor(k/2))), and Theorem A gives k >= (theta(x) - O(1))/(log G2 + O(1)), so the better the G2 bound assumed the higher the degree needed. | none | [attack-bonferroni-degree.md](history/staging/attack-bonferroni-degree.md) |\n| `Q-buchstab-deep-0830` | CLOSED | Can the certificate engine's deep-ladder Buchstab transfer be proven at dimension 2 at the depths the run levels reach by a route that does not go through the fundamental lemma at s >= 22.06 or the constant-free DH Theorem 9.1 (TODO 8a)? | NO by any instrument this corpus can cite, and the non-closing step is structural, not a constant: the transfer is an asymptotic for a ratio of two dimension-2 sifting functions at one sifting depth z = q, so a sieve bracket [f2, F2] survives undivided in the ratio (width 3.67 at the best sigma any level has, 4.71 at the @23 head); a correction \\|B - 1\\| >= 1% can only occur at sigma(q) < 1.8 where f2 = 0 and F2 >= 7.85, while a lower bound exists only at sigma > 4.266 where \\|B - 1\\| < 1e-5; the Buchstab identity iterated once is already negative at K = 1 for q = 37 at @23 and is the sieve itself when iterated fully; Jurkat-Richert fails Omega(1) as stated; in the certificate's legal direction F2 beats the trivial cap_K <= cap2 at 0 of 1512930 (q, K) pairs at @23, so the sharp-sieve floor is -1733138 at every K; the transfer stays HEURISTIC, what would move it is a kappa = 2 asymptotic in the open band 2 < sigma < 4.266, and part (b) is scoped and graded, not attacked. | 8 | [attack-0830-buchstab-deep.md](history/staging/attack-0830-buchstab-deep.md) |\n| `Q-buchstab-transfer` | ANSWERED | Is the certificate engine's Buchstab transfer provable in the shallow regime y = T^{o(1)}? | PROVEN conditional on the fundamental lemma and EMPTY at every run level (needs s >= 22.06; available 4.7 @23, 17.1 @97); the engine's B is a dimension-1 object, the survey's pair-Buchstab is dimension 2, the certificate needs a third. | 8 | [thm-buchstab-transfer-shallow.md](history/staging/thm-buchstab-transfer-shallow.md) |\n| `Q-bv-import-survey` | ANSWERED | Which unproven legs of the tile programme become theorems if Bombieri-Vinogradov, Elliott-Halberstam or GRH is imported? | Two legs become provable now, the fixed-modulus tail refinements S1/S2 by Siegel-Walfisz and the full-wheel prime-comb equidistribution in the tail regime by BV; the Buchstab transfer at bounded u and Assumption A for d = 2 are parity-blocked at every one of the four inputs. | none | [bv-import-survey.md](bv-import-survey.md) |\n| `Q-c2prime-drift` | PARTIAL | Why does c2' drift on the 22-term ladder? | Class-count-blind (attack-c2drift-01 §1); on item 1c's own band x >= 41 the ten-point rate is unresolvable: c2' 0.3745 +/- 0.1411 against c1's 0.3344 +/- 0.2358 on the same primes, a difference of 0.0401 = 0.12 of the control's ten-point window sd. Whether the drift is real is NOT closed here: attack-c2drift-01 measured it at p = 0.0053 on x >= 17 and left (i)' and (ii) alive. exponent-control §2/§4/§6 refit at 22 terms. | 1c (retired) | [attack-c2drift-01.md](history/staging/attack-c2drift-01.md), [c2prime-refit-22.md](history/staging/c2prime-refit-22.md) |\n| `Q-cal4-routing` | ANSWERED | Does the standing certificate plan point at the right obstruction, and what is CAL4 worth at @17? | The plan is aimed at the wrong obstruction, but not for partition W's reason: the @13 1e-9 gate is met today only by an assembly containing the brute-force C(990,4) march whose @17 counterpart is 513 days on ten cores and is not codeable here, the @17 requirement is a ladder whose first useful rung is relT4 <= 5.05e-5, a factor of 7.9 rather than six orders, and CAL4 has no correct constant value. | none | [phase1-W2-cal4-routing.md](history/staging/phase1-W2-cal4-routing.md) |\n| `Q-calm-lemma-identities` | PARTIAL | What in the fused window and the anchored calm is proven, exactly? | Three identities and the exact spectral forms are proven and verified as exact integers or floats; the uniform-in-q anticorrelation is REFUTED rather than open, with six counterexample primes across six levels, and the Anchored Typicality Measurement is the anchored-escape wall one level down with no proof mechanism in sight. | none | [natal-cap-19-calm-lemma.md](natal-cap-19-calm-lemma.md) |\n| `Q-calm-vs-kill` | PARTIAL | What does the anchored calm buy against annihilation? | Two theorems, the @11 closure and the X-Limitation Theorem at @11, @13, @17 and @19, and two refutations, calm implying concentration and minimizer loudness; the overlap-floor candidate that would prove non-annihilation dies on the anchor's own drift, with the anchored rotation t = 0 its worst case at @17, leaving the X-floor open. | none | [natal-cap-31-calm-vs-kill.md](natal-cap-31-calm-vs-kill.md) |\n| `Q-capK-bv` | ANSWERED | Is the prime-regime cap_K over the full wheel an unconditional asymptotic via the fundamental lemma over Bombieri-Vinogradov? | PROVEN at short-note grade for q_K = W^{o(1)}, and EMPTY in every computable range (needs s >= 10.82, first cleared at x = 263; the lemma's floor s >= 10 is first reached at x = 239); the deep ladder is untouched. | 11 (retired) | [thm-capK-bv.md](history/staging/thm-capK-bv.md) |\n| `Q-capture-identity` | ANSWERED | Does the transplanted cap family collapse onto a classical object, and what does the collapse say about the depth law and the wall? | PROVEN, elementary: floor_K = T − X(K) on the half-open window with both members in [Q², Q′²), a convention that is load-bearing; the wall is renamed, not weakened; Σ capU counts each member against its own threshold, not against a uniform p_K, the two differing 11% to 37% at K*. | Z2 | [quadpoint-identity-01.md](history/staging/quadpoint-identity-01.md) |\n| `Q-centered-discrepancy-estimate` | PARTIAL | Can the top of D_y be removed using ordinary prime distribution after an exact divisor flip, and what estimate remains? | Equations (3)-(4), the exact split and flip, survive review and finite controls. A repair of the top-range bound T^top=O_(A,eps)(x/log^A x) is DERIVED in section 3a (2026-09-08) with the endpoint atom retained, odd square divisors, the corrected reciprocal-totient local factors with (m',g)=1, explicit truncation powers, a Cauchy multiplicity device, and a uniform Mobius mean proved from the q=1 Siegel-Walfisz statements; its inputs are the terminal-point BV of Tao Notes 3 Theorem 17 with the prefix form derived by rounding. The handler (lane V) read the derivation on 2026-09-08, reconstructed (BV*), (3a.9), (3a.14) and (3a.16), checked the limiting constant of (3a.9) numerically at eight (b,g) pairs and found no defect; it is accepted at its stated scope, with ineffective constants. The payoff is D_y=D^(e_1)+O_(A,eps)(x/log^A x) only, so the handoff consumer is equivalent to D^(e_1)>=-4x/25+o(x). The fixed-endpoint signed discrepancy, D_y>=-4x/25+o(x) and twin-prime infinitude remain OPEN. | C | [centered-discrepancy-estimate.md](centered-discrepancy-estimate.md) |\n| `Q-centered-discrepancy-measurement` | ANSWERED | At reachable x, is the OPEN sufficient input D_y>=-4x/25 numerically violated, is the absolute form (13) numerically plausible, and does D_y carry structure beyond a random-sign model? | MEASURED to x=2^38. Neither pre-registered falsifier fires. D_y/x stays within 0.0043 of zero for the measured j>=26. Over 30<=j<=38, D_y/x differs from -(T1/x-C2) by at most 2e-4, with maximum 1.84401e-4 at j=30; the classical term dominates this finite comparison. The data do not establish an asymptotic fluctuation scale, an impossibility of informative future measurements, or the sufficient signed estimate. No proof status changes. | C | [centered-discrepancy-measurement.md](centered-discrepancy-measurement.md) |\n| `Q-certificate-engine` | PARTIAL | What did natal-cap-28 assemble, and at what calibration does the certificate engine stand? | There is no fully analytic certificate law and nothing here is closed: what reads as one object is nine at five calibrations with one of the nine refuted, the Certified-Head Theorem is PROVEN on 1/3/6 certified primes at @17/@19/@23, and every engine output carries the status of its weakest input, the HEURISTIC Buchstab Transfer Hypothesis. | none | [certificate-engine.md](certificate-engine.md) |\n| `Q-chain-review-0906` | ANSWERED | Do the written derivations from S(x) down to E_dagger hold as stated, with every imported theorem used inside its hypotheses and every discarded term at the claimed O_H(x/log^H x) rate, and where does the chain remain unreviewed? | Three independent adversarial passes found no defect in the six classical contributions from S(x) to E_>, none in the grouped moment (1)-(20) including the E_* to E_dagger step, and none in the residual-coverage link E_> to E_*, whose block-shape derivation was reconstructed and verified exactly. Every imported theorem was confirmed in its primary source except the Mobius Bombieri-Vinogradov citation, which points at an exercise sheet and needs a published replacement. Stale o(x) rates in three note bodies were replaced. A checked argument is not a refereed theorem; the OPEN signed margin is untouched. | C | [chain-review-0906.md](chain-review-0906.md) |\n| `Q-changelog-add-B` | ANSWERED | Which superseded text from wave-2 partition B belongs in the changelog? | The per-document entries for U-FRAME, ATTACKS3 and gate-multiplies, staged for the shepherd to merge in one pass; nothing was written to CHANGELOG.md here. | none | [changelog-add-B.md](history/staging/changelog-add-B.md) |\n| `Q-changelog-add-C` | ANSWERED | What history left wave-2 partition C's paper-suite files, and what has to be staged for the CHANGELOG? | Staged: Door 2's two load-bearing numbers are RETRACTED because the column they came from was disclaimed by its own script's correction banner, the corrected artifact reverses the conclusion, the OEIS term count is corrected to twelve, the matching lower bound is retired, and Zone Equivalence is demoted from spine theorem to framing device. | none | [changelog-add-C.md](history/staging/changelog-add-C.md) |\n| `Q-changelog-add-F` | ANSWERED | Which superseded text from wave-2 partition F, the natal-cap family, belongs in the changelog? | Entries for eleven natal-cap documents and two new files, led by the retirement of the Fused-Window Calm Lemma and the certificate law with it, plus the prose-note index that closes qc-arch E4; staged only, not merged. | none | [changelog-add-F.md](history/staging/changelog-add-F.md) |\n| `Q-changelog-add-G` | ANSWERED | What history left wave-3 partition G's files, and what has to be staged for the CHANGELOG? | Staged verbatim with file and line: the twin-candidate census index is corrected from A059861(n-1) to A059861(n) with four independent confirmations, two script banner titles are retired as contradicted by their own companion prose, and two scripts that had no banner title gained one. | none | [changelog-add-G.md](history/staging/changelog-add-G.md) |\n| `Q-changelog-add-P` | ANSWERED | Which superseded text from wave-4 partition P's three files belongs in the changelog, and what does calibration against the home document find? | Fourteen defects came out of the check wave 3 recorded it had not run, and none of them is reachable by any existing gate check; the removed or rewritten text is reproduced verbatim per file for the shepherd. | none | [changelog-add-P.md](history/staging/changelog-add-P.md) |\n| `Q-changelog-add-Q` | ANSWERED | What history left wave-4 partition Q's files, and what has to be staged for the CHANGELOG? | Staged: the Maier-Pomerance clause in the OEIS draft is RETIRED because it drew a >= from a << (they state an upper bound), and audit-numbers.js gained an oeis part of 39 checks carrying three retired values, including the exponent 1.57 quoted for G2 when 1.57 is A288815's. | none | [changelog-add-Q.md](history/staging/changelog-add-Q.md) |\n| `Q-changelog-wave2-A` | ANSWERED | What changelog entries does wave-2 partition A owe research/history/CHANGELOG.md? | Staged per-document entries for G2-STATE.md, LOCALIZED-GAP.md, maier-matrix.md and origin-excess.md, each naming the claim as it stood, what replaced it and the artifact that forced the change; a staging file for the parent to merge, not an edit to the changelog. | none | [changelog-add-A.md](history/staging/changelog-add-A.md) |\n| `Q-changelog-wave2-D` | ANSWERED | Which changelog entries does wave 2 partition D stage for CHANGELOG.md? | Staged entries in the changelog's house form for research/README.md (repurposed from a summary into the router), README.md, GLOSSARY.md, TODO.md, THE-DIALS.md, ZONE-POSTULATE.md, THE-LENS.md and FOLD-PROFILE.md, for the parent to merge. | none | [changelog-add-D.md](history/staging/changelog-add-D.md) |\n| `Q-changelog-wave2-E` | ANSWERED | What changelog entries does wave-2 partition E owe research/history/CHANGELOG.md? | Staged entries covering covering-dive.md, two-class-lower-bounds.md, PRIOR-ART.md, sift-limit-attack.md, theta-ladder.md, maxgap-law.md and dhr-verification.md, led by the FKMPT constant moved across two files together. | none | [changelog-add-E.md](history/staging/changelog-add-E.md) |\n| `Q-changelog-wave3-H` | ANSWERED | Which changelog entries does wave 3 partition H stage for CHANGELOG.md? | Entries for U-FRAME.md's split, in which sections 10, 11 and 12 keep their numbers and heading text and lose their bodies, plus the three new child documents kappa-not-L.md, operator-and-pair-count.md and f-decays.md. | none | [changelog-add-H.md](history/staging/changelog-add-H.md) |\n| `Q-changelog-wave3-I` | ANSWERED | What changelog entries does wave-3 partition I, the flagship split, owe research/history/CHANGELOG.md? | Staged entries for the split of the flagship's wall chapter into the new paper/wall-note.md and the pointer repairs that split forces, in house form, newest first and grouped by the document changed. | none | [changelog-add-I.md](history/staging/changelog-add-I.md) |\n| `Q-chen-fold-benchmark` | ANSWERED | Can the fold framework reproduce a classical Chen lower bound on long intervals with every imported hypothesis and error budget stated? | The classical implication is derived below using three explicitly imported theorem statements, with the combinatorial minorant proved and its integral margin certified by exact rational arithmetic. This is a benchmark, not a new prime theorem, and it does not prove twins or a short-zone result. | C | [chen-fold-benchmark.md](chen-fold-benchmark.md) |\n| `Q-chen-opportunity` | ANSWERED | Does the signed Chen target identify an approachable proof mechanism, and what should the next research obligation be? | The target is sufficient but subtracts an avoidable odd-composite penalty; even cancellation of its obvious signed components does not close the existing separate bounds. Select a prime-detecting weight against explicitly justified arithmetic input ranges before pursuing a new estimate. The opportunity is a research judgment, not a proved mechanism. | C | [chen-opportunity-audit.md](chen-opportunity-audit.md) |\n| `Q-chen-signed-target` | PARTIAL | What exact additional arithmetic estimate would turn the completed Chen benchmark into a proof of infinitely many twin primes? | An elementary prime minorant gives a conditional twin theorem with all weights, ranges, quantifiers and errors specified. The opportunity audit identifies an avoidable negative-weight penalty and a separate-estimate budget that fails even under component cancellation; this sufficient target remains OPEN and is not the sole next proof obligation. | C | [chen-signed-target.md](chen-signed-target.md) |\n| `Q-coefficient-structure` | ANSWERED | Does the arithmetic structure of the aggregated coefficients reduce the endpoint estimate beyond the previous rectangle, and can smaller coefficient norms alone supply the remaining power saving? | Exact classification gives a nonsquarefree L2 norm of size at most sqrt(D)*W^(1/4) times logarithms, while the squarefree top-band energy is at least a constant times D*W*log(W). At d~x^0.277, e~x^0.467 every endpoint sector except the squarefree-squarefree prime sum has a power saving. This reduces that rectangle to one explicit endpoint correlation; prime-dispersion.md supplies the remaining estimate and controls the full rectangle. Finite algebra, CRT sectors and archived prefixes are checked separately. | C | [coefficient-structure.md](coefficient-structure.md) |\n| `Q-cofactor-progression-transfer` | PARTIAL | Can the corrected full-cofactor kernel be estimated on a growing cofactor range by representing its actual arithmetic on CRT progressions, and does the resulting bound reach the twin consumer? | Derived from named imports: for some positive absolute kappa and d, all prime-r cofactor pairs s,t <= floor((log x)^kappa), including non-squarefree inputs and mixed transition/smoothed profiles, admit a dyadic scale-average O(log^(2-d) X) bound after division by x. The extension uses bounded multiplicative Euler modifications on the progression s\\|n, t\\|n-2 and keeps its density 1/lcm(s,t), so summing cofactors costs only O((log log X)^2). This goes beyond s=t=1 but remains short of o(x). The large-cofactor joint remainder, outside residual and twin margin remain OPEN; the attempted extension to power-sized cofactors fails on source modulus range and accumulated logarithmic cost. | C | [cofactor-progression-transfer.md](cofactor-progression-transfer.md) |\n| `Q-coherence-0828` | ANSWERED | Are the 2026-08-28 notes consistent with each other? | 20 contradictions, 8 convergences, 46 live-doc corrections collected (34 mechanical, 12 need a second reader). | none | [coherence-0828.md](history/staging/coherence-0828.md) |\n| `Q-coherence-0829` | ANSWERED | After the 2026-08-29 edits, does every live document state each changed fact the same way, and do the papers still agree with the notes? | 24 changed facts checked at 63 live-layer sites: 16 live disagreements (11 APPLY, 5 HOLD) plus one count error inside a HELD note; the worst is the localized band 2.9 to 4.2 attributed to \"two engines\" when only one of the two reads it, the other reading 2.14 to 4.39; qc.js 1 finding, in a file another agent holds; audit-numbers 251/251. | none | [coherence-0829.md](history/staging/coherence-0829.md) |\n| `Q-coherence-0830` | ANSWERED | Do the live layer (TODO, README, G2-STATE, REFUTED, anchored-calm, GLOSSARY, CHANGELOG, QUESTIONS) and the eighteen notes landed 2026-08-29 evening and 2026-08-30 morning agree sentence by sentence? | They mostly agree; 27 distinct disagreements found, 15 graded APPLY (both sides quoted, no judgement needed) and 12 ASK, none touching a note's mathematics: the live layer lags the landed notes in six places (a PROPOSAL listed under PROVEN in G2-STATE section 0, 67 for 71 closed routes, the skeleton door still open in README and two anchored-calm sentences, Z5's body contradicting the record-mechanism note, item 9's one-versus-two open steps, a stale Z2 first move), carries four numeric slips (8-of-10 for 10-of-10, 27 for 25 clauses, u* in the seventh digit, nine for ten levels), one rung inflation (a text-layer reading called read at the page), and one landed note (recon-0830-rec-killrun) is itself wrong about the record on the seventy-five count; qc.js --full at 09:23 fails only on two ledger ids from notes that landed during the run. | none | [coherence-0830.md](history/staging/coherence-0830.md) |\n| `Q-comb-discrepancy` | PARTIAL | Can the Comb Discrepancy Lemma's constant 2*3^k be tightened (TODO 8c)? | The lemma IS the trivial per-block bound; the attained optimum is max G - min G, a computed number per level (slack 29x @23, 54x @29, growing); term-by-term pricing lifts the certified head share to about 28-29 percent, not the 50 hoped for. | 8 | [comb-discrepancy-tight.md](history/staging/comb-discrepancy-tight.md) |\n| `Q-comb-tail-0830` | PARTIAL | Can TODO item 8's remaining part (b), prime-comb equidistribution in the tail regime, be proven at the depths the certificate needs, or at least measured and priced at the enumerable levels (8b)? | Not proven and no exponent moves: the tail defect is measured exactly for the first time (signed -0.01% of the tail main term at @23, -0.00% at @29, unsigned 0.60% and 0.25%, smaller than the head's 1.00% in aggregate but 8 to 37% per prime at its worst), the Comb Discrepancy Lemma's method fails in the tail on its MAIN TERM (off by a derived factor e^gamma(1/u - q^(1-u)), measured aggregate 1.14x at @23) before its error term (10^27 x main), the prime-comb Legendre form has 2^(k+2) terms of which the short ones obey the lemma's <1 bound and the long ones are prime-in-AP discrepancies whose term-by-term sum is 11.2% of main at @23 against a signed -0.01%; the all-x form at K = 0 is Theorem C (PROVEN, limit only) and at the certificate's depth K* the non-closing step is a kappa = 1 asymptotic at sifting parameter s < 2, a proof gap on data that show B_tail = 0.998-1.000 at K* through @29; the full-depth half is 8(a), CLOSED. | 8 | [attack-0830-comb-tail.md](history/staging/attack-0830-comb-tail.md) |\n| `Q-consolidation-wave` | ANSWERED | Can the five cost-class scripts be bound to their own output, and can U-FRAME's two lost tables be rebuilt? | All four runnable cost-class embeds are bound and the hand-pasted-tail counter falls 19 to 15 with the gate at TOTAL 0; three needed --force at a cost of 24 tokens out of 211 with no value changed anywhere, and natal-cap-33-overnight.js is a three-run composite that should never have been on the cost list. | 11b (retired) | [consolidation-wave.md](history/staging/consolidation-wave.md) |\n| `Q-constants-audit-2` | ANSWERED | For every load-bearing number that appears in two or more live documents, does it agree everywhere, and does it trace to exactly one custody-bound producer? | 25 findings over 69 of 70 files, 8 rank A, 12 rank B, 5 rank C: 46 named-constant groups carry two or more distinct values across two or more files, and 439 numerals of five or more significant digits in the live layer have no producer printing them at any precision; nothing was fixed and five judgement calls stayed queued. | none | [constants-audit-2.md](history/staging/constants-audit-2.md) |\n| `Q-consumer-comparison` | ANSWERED | How does the sufficient margin C_2 x + E_dagger(x) >= c_0 x/(log x)^K compare logically with the published sufficient routes to twin primes, and what does the literature already own of the corner object? | The sufficient margin remains OPEN. The published conditional routes imply it where their hypotheses and stated lower bounds apply; reverse implications are not established. The dictionary is E_dagger(x)/x = C_2(2 delta_x - delta_{x/2} - 1) + o(1), with K>0 below that normalisation's resolution. Independent review corrects the comparison: cumulative and dyadic consumers on unbounded scales are equivalent at the same fixed K with possibly different positive constants. At K=0 an eventual cumulative lower bound also implies the unbounded-scale dyadic consumer. The nonnegative-weight corner identification applies only to the prime-cofactor subfamily; the full residual retains additional branches. Every surveyed route remains conditional. | C | [consumer-comparison.md](consumer-comparison.md) |\n| `Q-corner-branch-diagnostic` | ANSWERED | Can future experiments enumerate the full coefficients C(n) and C'(n-2) branch by branch, without silently dropping proper prime powers, cofactors s>1, non-squarefree inputs or either gcd branch? | The driver exists and its two independent constructions of each side agree exactly on 8192 retained sides and 1802 fixture sides; five deletion and reconstruction controls are ACTIVE in both runs. Its split by gcd(n,n-2) measures integer parity, not the per-term CRT gcd partition. The fixed-eta corner is empty on the retained prefix, and Elo exceeds E0 across the entire J_x, so the matrix is a proxy window. This is finite algebra and tool preparation, with no asymptotic estimate or twin margin. | C | [corner-branch-diagnostic.md](corner-branch-diagnostic.md) |\n| `Q-corner-coefficient-energy` | ANSWERED | Can the full corner coefficients be studied before splitting their cofactor branches, and what do exact energy and classical smoothed sieve weights supply? | The sharp coefficient energy is an explicit signed quadratic form in short cofactors plus O_epsilon(x^(1/2)K^2*x^epsilon), a power below x at the fixed corner. Logarithmically averaging the divisor cutoff gives Barban–Vehov weights and an O_eta(x log x) squared norm on each side and absolute shifted-product bound. The follow-up sharp-corner-transition.md gives sharp O_eta(x log^2 x) norms and proves a non-negligible norm transition for sufficiently small fixed eta. Signed transition saving and the global complement remain OPEN; no exact residual cut or twin margin changes. | C | [corner-coefficient-energy.md](corner-coefficient-energy.md) |\n| `Q-corner-correlation` | PARTIAL | On the corner S_0 where both cofactors sit just above their own cutoffs, what exactly is the remaining endpoint sum, what does the sufficient consumer need there, and does any reviewed correlation theorem supply it? | The full corner is exactly sum_n C(n)C'(n-2). Proper-prime-power terms above the cutoffs are negligible at fixed eta by §1.1; the remaining prime-r terms with s>1 or s'>1 and non-squarefree inputs remain OPEN. The s=s'=1 piece has a nonnegative Mobius-product weight with exact moving cuts. The multiplicative lift and round-review sampling give small continuous and dyadic scale-average log savings for that subfamily, with no o(x), full-corner control or twin margin. The complement is still open. | C | [corner-correlation.md](corner-correlation.md) |\n| `Q-corner-log-average` | PARTIAL | What do logarithmically averaged correlation theorems and their quantitative successors actually supply for the prime-cofactor corner weight, with exact support, rates and scale quantifiers? | The pure band is invariant under dilation by primes outside it; the exact cofactor window is not, but its support extends to x^(1-2eta) and need not leave the window under a fixed dilation. Uniform-prefix logarithmic bounds imply block bounds by Abel summation; a relative exponent above 3 is one sufficient route to o(x), not a universal necessary threshold. Pilatte's 1/(96e) calculation applies only to his displayed parameter choice. A bounded multiplicative Fourier lift now gives a small continuous scale-average log saving for the prime-cofactor subfamily via Tao-Teravainen v2; see prime-band-transfer.md. No o(x), dyadic pointwise estimate, full-corner estimate or positive twin margin follows. | C | [corner-log-average.md](corner-log-average.md) |\n| `Q-corner-measurement` | PARTIAL | At finite dyadic x, how large is the corner two-point correlation K(x)=sum_{x/2= 1 with the density condition as its only hypothesis, so what beta2-note adds is two parameter choices and one dimension-free observation that is zeb's at kappa = 1; the honest response is to state the k-class family, whose k = 1 row is a known theorem. | none | [attack-dim2-standalone.md](history/staging/attack-dim2-standalone.md) |\n| `Q-discrepancy-two-class` | ANSWERED | What is the two-class analogue of Holt's signed discrepancy DeltaPhi, and do the three candidate growth laws on the table agree? | One object in three norms rather than three agreeing measurements: the proven Level Ledger bound 2*3^(pi(x)-1) and natal-cap-29's spectral level law share the integer 3 provably, since R_k(p) tends to k+1, so the apparent agreement is an identity; the custody gate reproducing Holt's Table 2 passed. | none | [discrepancy-two-class.md](discrepancy-two-class.md) |\n| `Q-dispersion-range` | ANSWERED | How far do the dispersion and coefficient-sector estimates extend across divisor sizes, and does a better orientation or gcd budget enlarge their overlap? | Averaging the nonzero gcd losses over the original divisor gives a different dispersion bound with cross-prime exponent (5/4)a+(1/2)b+3/50, where a=delta+6/25 and b=nu+1/20. Explicit diagonal, prime-cut and sparse-sector inequalities control a region of full rectangles, including d~x^0.282, e~x^0.522 at product scale x^0.804. Its cross-prime and left-sparse budgets are 0.9985 and 0.9988. The supremum of product exponents in this sufficient region is 5397/6700; it is not a uniform product cutoff. Balanced supports separately allow 12/25= pi(2s)-pi(s) (K* = 17 at s = 16 by exact walk, VERIFIED, so the certificate reads 18 against 8 on the chain itself; it exits the whole legal band at s = 128, PROVEN); the sharper maxsum bridge Ghat(2s) <= maxsum_{K*+1}(T_s) is PROVEN and holds under 8 at all fourteen enumerable steps (VERIFIED), but its all-s form needs an upper bound on K* against the entering primes' two-class covering that nothing proven supplies; the doubling inequality itself is untouched. | D | [attack-0829n-doubling-bridge.md](history/staging/attack-0829n-doubling-bridge.md) |\n| `Q-doubling-killrun-0830` | PARTIAL | Can the exact per-fold L bounds of the pi(2s) - pi(s) folds inside one doubling step compose to an upper bound on the weighted kill-run below the allowance 8 Ghat(s)/gbar(s), for all s, without passing through K*? | No, and the composition is closed as a route, not merely unproven: the folds compose as a PRODUCT, K*+1 <= prod_j (1+L_j) (PROVEN here, exact at fourteen steps, 540 against 18 at s = 16), the composed certificate never sits below yesterday's maxsum certificate (PROVEN by monotonicity), it exceeds the allowance at 8 of 14 enumerable steps starting at s = 9 (VERIFIED), and since every fold kills a slot the composed index is at least 2^N, whose floor 2^N gbar(s) alone exceeds 8 Ghat(s) at the chain rungs 16, 32, 64 and beats the cited polynomial ceiling on Ghat for all large s (PROVEN given the ceiling); the sum form is false at seven steps and the max form is a floor; the weighted run itself, computed exactly, sits at 0.83 of its allowance at s = 16 and (M8) is exactly where yesterday left it, OPEN. | D | [attack-0830-doubling-killrun.md](history/staging/attack-0830-doubling-killrun.md) |\n| `Q-embed-backlog-0829` | PARTIAL | How much of the embed backlog (legacy scripts with no OUTPUT block, readings not traceable to their block) could be closed in one pass, and what did the fresh runs contradict? | MEASURED. embed-backlog fell 218 -> 189 findings in one pass: no-output-block 34 -> 5, readings-not-traceable 181 -> 181 (untouched by design), hand-pasted-tail 3 -> 3. Twenty-nine legacy scripts were bound with node research/qc/embed.js, zero --force was used, and one legacy reading is contradicted by its own fresh output (attack-beta2-03-exact-strata reading 8, \"1.04 to 1.07 across y = 7..23\" against a printed 1.102 at y = 7). | none | [embed-backlog-0829.md](history/staging/embed-backlog-0829.md) |\n| `Q-embed-binding` | ANSWERED | What does the embed fingerprint actually bind, and what does it leave loose? | It binds the code to a claim about the output and does not bind the output block to anything: code-sha256 covers every byte above the banner, out-sha256 is only ever compared against another run, timeouts and failures write nothing, a prose line can unbind a tail and one did during this pass, and 37 cited scripts have no output custody at all. | none | [verify-the-verifier-embeds.md](history/staging/verify-the-verifier-embeds.md) |\n| `Q-endpoint-fourier` | ANSWERED | Does a complete Fourier truncation budget and a matched Kloosterman-fraction theorem control any further part of the actual endpoint sum when its small prime-power coefficients are combined first? | A written classical-input derivation controls the rectangle d in (floor(x^(27/100)),2floor(x^(27/100))], e in (floor(x^(46/100)),2floor(x^(46/100))] to O_H(x/log^H x) for every fixed H. Its product is of order x^(73/100), outside the previously controlled region eventually. The aggregated Fourier exponent is 1989/2000; the full Vaaler tail and gcd=2 branch are included. This is a regional estimate, not a full residual bound or twin theorem; finite algebra and saved-factor checks are separate validation. | C | [endpoint-fourier.md](endpoint-fourier.md) |\n| `Q-endpoint-pairing` | ANSWERED | Does retaining the difference of interval endpoints improve the complete Fourier budget, and are the lowest frequencies actually the next obstruction on the squarefree pilot rectangle? | A written derivation controls the full rectangle d~x^(11/40), e~x^(93/200) to O_H(x/log^H x), with product of order x^(37/50). The sufficient first exponent becomes 3/20+(7/10)(a+b)+(1/4)max(a,b), equal to 3999/4000 there. On the a=b=517/1000 pilot, this transition-frequency budget is 20061/20000, not the separate-endpoint 20163/20000; prime-dispersion.md now controls that pilot with a different estimate. Finite identities and exact exponent checks validate the implementation, not the asymptotic theorem. No uniform product cutoff or twin lower bound follows. | C | [endpoint-pairing.md](endpoint-pairing.md) |\n| `Q-endpoint-target-audit` | ANSWERED | Does the endpoint reduction require a fixed positive fraction of the expected twin count, and do recorded uniform-gap theorem failures exclude its weaker consumer? | No fixed fraction is required: for every fixed H the reduction has error O_H(x/log^H x), so C2*x+E_>(x)>=c*x/log^K x on unbounded dyadic scales suffices, as does a stated logarithmically rescaled average. These implications are derived from named inputs; their endpoint hypotheses remain OPEN. The recorded uniform-gap theorem comparison has different quantifiers and does not establish an obstruction for this consumer. | C | [endpoint-target-audit.md](endpoint-target-audit.md) |\n| `Q-erdos-map` | ANSWERED | Does the two-class object already occupy a numbered slot on somebody else's map of open problems? | It has no problem number and no name on erdosproblems.com, Ben Green's list or the OEIS citation graph, but it is not unposed: Erdos posed it himself in one sentence in the same paragraph as the $1000 offer, and our object is the r = 2, difference-2 instance of his own stated extension of #687. | none | [attack-erdos-map.md](history/staging/attack-erdos-map.md) |\n| `Q-excess-chain-c` | CLOSED | Does the excess chain's constant c pin to 1.074 via a compressed-tail max correction? | NO: for any FIXED correlation length the derived family tends to 1 and both named corrections lower the maximum, so 1.074 is not derivable from a compressed-tail max correction; c measures 1.05 +/- 0.07 (sd) over x = 17..31, MEASURED, which does not separate 1.05 from 1.074. | 10 (retired) | [excess-chain-c.md](history/staging/excess-chain-c.md) |\n| `Q-exponent-control` | PARTIAL | What is the growth exponent of G2, measured against a control that knows its own answer? | MEASURED 1.50 +/- 0.05 stat on 22 trusted terms after control correction, systematic unquantified; the raw fit's bias grows and sticks at +0.28, exponent 2 stays disfavoured but is not excluded, and the question is bias-limited rather than data-limited. | none | [exponent-control.md](exponent-control.md) |\n| `Q-external-data-audit` | ANSWERED | Where is the corpus limited by external data it could simply adopt? | Findings are candidates, not verdicts: the highest settles an open reading for free, since dividing the two now-trusted series moves the G2/h ladder maximum from x = 37 to x = 79 and reverses the fall at x = 53, killing \"peaks at x = 37\"; the rest tighten quoted constants or retire stale limitation sentences. | none | [external-data-audit.md](history/staging/external-data-audit.md) |\n| `Q-f-census` | PARTIAL | What does the grain census law give for f, the qualifying-gap fraction, and where does it structurally stop? | DEFECT-carrying: every census point from x = 37 up is aliased, so the 42-point table above x = 31, both regressions, the factor 170 and the x = 1000 extrapolation must be re-read against the corrected law; the staircase theorem (treads are exactly the twin pairs), Lemma B, the comb coefficient and everything at x <= 31 are UNAFFECTED, and the growth law is a fit and not a theorem. | none | [a3-03-f-from-census.md](a3-03-f-from-census.md), [f-decays.md](f-decays.md) |\n| `Q-f4weak` | PARTIAL | What is the weakest sufficient form of F4 that could be proven, given rms <= z^(3+o(1))? | The mode-count/degree bound is NOT the theorem (sqrt(N)*rms grows at z^8.174 +/- 0.249 against the z^4.26645 ceiling and is dead from z = 31); one new PROVEN finite-z clearing survives z = 47 at a 5.0 percent margin but converges onto the trivial bound; the weakest law-shaped form found is MV(alpha), a mean value over modes, MEASURED slope 4.214 +/- 0.304 with beta2 inside the bar. | none | [attack-f4weak-01.md](history/staging/attack-f4weak-01.md) |\n| `Q-fdecay-deep` | ANSWERED | Does the fitted f decay law hold out of sample at deep levels? | It holds as a band, nine of nine deep levels inside the four-specification band, but the residual trends against the central specification at t = -7.08; the test also found what it was not looking for, that the 42-point exact census is wrong from x = 37 upward by a factor rising to 1.63 at x = 199, from a 32-bit shift alias. | none | [fdecay-deep.md](history/staging/fdecay-deep.md) |\n| `Q-fdecay-out-of-sample` | OPEN | Does the 42-point decay law for f, the qualifying-gap fraction, survive out of sample, given that every downstream reading extrapolates it three to thirty times past its range? | Pre-registration only, written before any producer for the pass exists on disk: four specifications are refit on the 42 exact census points as a transcription check and their projections are frozen, and nothing here is measured. | none | [fdecay-deep-prereg.md](history/staging/fdecay-deep-prereg.md) |\n| `Q-fekete-1d-defect47` | PARTIAL | Does the bounded-defect Fekete route survive a probe at 47#? | The defect at 47# is ordinary and the 47# enumeration cannot move item 1d's TPC threshold whatever value it returns; the bounded-defect Fekete lemma is stated exactly and proved with every hypothesis except the candidate itself discharged, so the route reduces to one named inequality and is neither closed nor open beyond that. | 1d | [fekete-1d.md](history/staging/fekete-1d.md) |\n| `Q-fixed-endpoint-discrepancy` | PARTIAL | After the accepted truncation to odd moduli e=-4x/25+o(x)? | Reviewed 2026-09-09 after the coprimality repair: T_I^low=O_(A,eps')(x/log^A x), hence S=C_2x+B+O_A(x/log^A x), with B the exact Type II plus band sum (2.9). The untruncated modulus bound was false; g<=(log x)^(A+13), both tails, the coprime density and the weighted BV multiplicity now pay the estimate. Review corrects harmless log powers and the power-of-two atom over the full eps' range. D^(e_1)>=-4x/25+o(x) is equivalent to B+2C_2M>=-4x/25+o(x); it implies, but is not equivalent to, B>=-(C_2-1/200)x+o(x). The latter and H_B are sufficient OPEN margins. The absolute band statement (4.9) is one stronger sufficient input, not a necessary condition. No inspected source estimates the remaining actual signed B. Twin-prime infinitude remains OPEN. | C | [fixed-endpoint-discrepancy.md](fixed-endpoint-discrepancy.md) |\n| `Q-fkmpt-corrigendum` | ANSWERED | Does the 2023 FKMPT corrigendum change anything the corpus depends on? | Clean bill of health: the corrigendum changes four numerical constants and the parameter M, every one already carried at its corrected value here; the loudest finding is the opposite of the expected one, since the premise that nothing in this repository has read it is false and has been since 2026-08-18. | none | [verify-fkmpt-corrigendum.md](history/staging/verify-fkmpt-corrigendum.md) |\n| `Q-fold-arithmetic-bridge` | PARTIAL | Does one parity-table bridge from the anchored fold ledger yield a sufficient twin lower bound with named arithmetic inputs? | Exact identities retained. The one unread sieve input of the pricing, Bombieri--Vinogradov for k-fold X^(1/u)-rough products, is derived from Wu's Lemma 2.3 (section 3a); the same input, sieved in the composite variable, gives the contamination aggregate constant 4 for each fixed k,u, replacing the displayed 12.86 to 19.72. With these inputs, elementary bounds give Q_cov(u)<1 and c*_real(u)<4 for every u>4 (section 4a, independently reviewed 2026-09-09 with rational certificates), so neither sufficient ratio test succeeds at any depth. This closes the two tests, not the decorrelation hypotheses, and supplies no twin estimate. | C | [fold-arithmetic-bridge.md](fold-arithmetic-bridge.md) |\n| `Q-fold-profile` | ANSWERED | Where does a fold by p land its damage inside the original tile stretch [0, W)? | The counting half is a theorem with an effective constant (Level Ledger PROVEN, Mirror Ledger VERIFIED 45 of 45) and the survival law is scale free (MEASURED), but the quantity controlled here is provably uninformative about L and kappa(m), so it buys nothing against the Zone Postulate route. | none | [FOLD-PROFILE.md](FOLD-PROFILE.md) |\n| `Q-fold-succession` | CLOSED | Does folding in succession damp the multiplier, giving a lower L than the sum of per-fold jumps? | No: the succession-damping route closes empirically at both reachable scales against a calibrated null, the only damping present is what level-tracking forces and the only extra structure present is cooperative, matching the structural closure by super-additivity. | none | [fold-succession-autocorr.md](history/staging/fold-succession-autocorr.md) |\n| `Q-foldL-amortized` | CLOSED | Can the fold multiplier L be bounded by charging kill-runs to the birth of the gaps they consume? | The accounting does not close on the tile: the Consumption Identity makes the charge exact and the bound is then only as good as X = 2Q0 + Q+ + Q-, and supply against demand fails on the tile at every x by exactly the factor ln^2 x; it closes in a localized window from p about 421 on, which is a statement about a bounded population only. | none | [attack-foldL-04-amortized.md](history/staging/attack-foldL-04-amortized.md) |\n| `Q-foldL-bridge` | ANSWERED | Does the bridge from a fold's kill count to L hold, and what is the sharpest transportable ceiling it gives? | The bridge is PROVEN and was already in the corpus, so the job was custody rather than a new theorem; the new pair-window rung is exact at all eight diagonal cells, and the best fully-proven transportable ceiling stays near 0.18p and beats the requirement nowhere. | none | [attack-foldL-02-bridge.md](history/staging/attack-foldL-02-bridge.md) |\n| `Q-foldL-extinction-scaling` | ANSWERED | Does the multi-kill extinction law predict windows it was not fitted on? | CONFIRMED under the pre-registered rule at Y = 2e7, 2e8 and 2e10 from the Y = 2e9 calibration alone, with the run-length extension REFUTED exactly as pre-registered; this promotes a law candidate on measurement, not a proof. | none | [attack-foldL-06-scaling.md](history/staging/attack-foldL-06-scaling.md) |\n| `Q-foldL-maxsum-direct` | CLOSED | Can maxsum_m(T_x), plain and residue-deleted, be bounded above by an argument that never counts kills? | The brief's premise contradicts the corpus's own Bridge Floor (maxsum_k >= maxsum_1 = G2(T_x) for every k), so the bridge cannot convert far enough however good the bound is; all four routes close (R1 partial then closed, R2 tautological and quantitative, R3 busts the budget), and the order-m object is in print, so item 0c's absence sentence has to change. | 0c | [attack-foldL-05-maxsum-direct.md](history/staging/attack-foldL-05-maxsum-direct.md) |\n| `Q-foldL-run-census` | PARTIAL | What ceiling on L does the channel-compatible run census force, and is that ceiling forced or a bridge? | Channel compatibility is exact and cheap but buys little: 3 in 71 at kappa(1) = L, the forced ceiling is a presence-relaxation rather than a multiset-exact bound, exact tiles stop at T_29, and the angle's own verdict sends it to TODO 0c, bounding a residue-deleted maxsum, which this pass did not attempt. | none | [attack-foldL-01-census.md](history/staging/attack-foldL-01-census.md) |\n| `Q-foldL-transport` | ANSWERED | Is there a third invariant of the tile's gap alphabet that provably survives the fold recursion, alongside the sum rules and the maxsum family? | Yes: the tail-count profile transports (the Tail-Count Transport, PROVEN, verified to T29 at 214,708,725 twin slots) and is sharper per fold than the standing route by a factor of three to eight in nats, but it does not chain, dying on the same accumulating index gate-multiplies.md closes. | none | [attack-foldL-03-transport.md](history/staging/attack-foldL-03-transport.md) |\n| `Q-foldL-window5` | MIXED (foldL-window5-prereg.md: OPEN; foldL-window5.md: ANSWERED) | Does the fold extinction law score one decade blind at W = 2e11? | HIT on all four pre-registered criteria (last fold with L >= 2 measured 631 against band [571, 877]; both counts inside their bands), with the hot systematic now confirmed at five of five windows at measured over predicted about 0.82, which by the prereg's own clause belongs in the law as a known overprediction rather than in the acceptance band. | none | [foldL-window5-prereg.md](history/staging/foldL-window5-prereg.md), [foldL-window5.md](history/staging/foldL-window5.md) |\n| `Q-ford-halberstam` | CLOSED | Does Ford and Halberstam's dual decomposition beat Brudern-Fouvry's (2.6), as its authors said it should? | No: the dual is the same inequality, verified to 3.553e-15 at r = 2 and 6.821e-13 at r = 2..7, and their estimation step is better than they claimed, the Rosser boundary layer making their absolute-value bound an identity with loss 0.000000000 at z = 20, 30, 42; the shortfall is the freedom the step gives up, one level per component and no lower sieve. | none | [attack-ford-halberstam.md](history/staging/attack-ford-halberstam.md) |\n| `Q-fractional-retention` | CLOSED | Do Brady and Li's fractional retention rules at kappa = 2 open the adjacent door sift-limit-attack section 7c named? | Do not pursue: the door is real, applies to our system and carries no regress, but half the rule family cannot be applied at kappa = 2 and the applicability window there is empty, both settled from the sources; the unlooked-for find is that Brady's thesis is prior art for our object. | none | [scope-fractional-retention.md](history/staging/scope-fractional-retention.md) |\n| `Q-frontier37` | ANSWERED | What do the fold-L wave's three exact instruments give one level deeper, at fold 37? | The prior-art check moved the target first: the ninth diagonal L(T_31, 37) = 4 and the maxsum_m(T_31) table for m <= 8 were already embedded in the corpus, so this is a re-measurement, not a first; M_alt = G2 is measured at eight folds and proven at none, and the transport ratio's rise 0.8881 to 0.9477 has no mechanism attached. | none | [frontier37.md](history/staging/frontier37.md) |\n| `Q-full-coefficient-average` | PARTIAL | Can aggregating the complete coefficients before a correlation theorem remove the explicit cofactor count, and what additional estimate is needed? | Exact factor and rounded-endpoint Fourier identities retained, with c_(i,0)=3/5 and an explicit composite-filtered weighted sum. The full family is not 1-bounded, but the sufficient phase condition admits at least k=0,+/-1 on the left and l=0,+/-1,...,+/-6 on the right for every Mellin twist; the earlier zero-only claim is corrected. Composite filtering and the required correlation rate remain unmatched. Proposition 6.5, independently reviewed including on 2026-09-09, proves coefficient norm at least (log x)^(2/5) for representations by 1-bounded functions on all smooth inputs. This does not exclude density-one representations, paid growing components or a jointly treated Fourier sum. No sufficient signed twin margin follows. | C | [full-coefficient-average.md](full-coefficient-average.md) |\n| `Q-g2-43-term` | ANSWERED | What is the fourteenth exact term of the G2 ladder, past 41#? | G2(43#) = 618, at least position 830,330,079,152,051 and attained at 8 positions of the period, confirmed by two independent enumerations over disjoint natal masks agreeing on value, multiplicity, position and survivor count; 47# is priced by direct probe at 1.49 days per run and stays out of session reach. | none | [phase1-T2b-exact-ladder.md](history/staging/phase1-T2b-exact-ladder.md) |\n| `Q-g2-falls-decision-rule` | PARTIAL | Does G2(x#)/x^2 fall, and what pre-registered decision rule would settle it? | The instrument is calibrated and sharp, resolving the one-class control at 11.6 sigma on the same nine points where it returns 0.9 sigma on G2, and the honest band on the slope is +/-0.117 at nine terms, so the rule is registered; but the power analysis puts separation at x = 53 only if the falling law is x ln^2 x and at x = 151 if it is x ln^3 x, so phase 1's extra terms settle nothing and no reachable exact ladder does either. | 1d | [phase1-T3prep-decision-rule.md](history/staging/phase1-T3prep-decision-rule.md) |\n| `Q-g2-provenance-x37` | ANSWERED | Does the headline G2 exponent depend on the eight terms this repository has never reproduced, and does any instrument without G2(37#) inside it flag x = 37? | MEASURED, both largely negative: the custody-only refit lands at 1.533 against the headline 1.498 and no provenance-block drop moves the corrected central by more than 0.035 (Ghat(64) alone moves it 0.002), while no G2-free instrument flags 37 in the high direction, the one \\|z\\| past 2 being h(37#) at -2.20 in the low direction and range-dependent; the blind extreme-value forecast of G2(37#) does overshoot at z = +6.58, unexplained. | 0 | [measure-g2-provenance-0829.md](history/staging/measure-g2-provenance-0829.md) |\n| `Q-g2-state` | PARTIAL | What is known about G2, the two-class maximum twin-slot gap, and at what calibration? | The gap is unchanged through roughly fifty recorded passes: the proven upper bound is exponent 4.26645 against a target of 2, and this file consolidates the map rather than computing anything new. | none | [G2-STATE.md](G2-STATE.md) |\n| `Q-gap-spectrum` | ANSWERED | What is the full gap-length distribution of twin slots over one period of the tile, what does the excess functional \\|B_N\\| = sum_{g>N}(g-N) read on an x^s grid, and where does the empirical minimum window count leave zero? | The distribution is exact at nine levels to x = 31 and the pre-registered renewal null FAILS in the light direction at every one: the tail is uniformly steeper than exponential (body slope -1.3302 against -1), no power law survives (the power-law slope moves by a factor 17), the maximum is 2.087 to 2.241 BELOW the renewal maximum, and \\|B_N\\| sits at 0.0245 to 0.7436 of the null and is exactly zero at s = 1.75 where the null predicts 6.5262e+5 empty windows; so no heavy-tail mechanism for a super-linear G2 exponent exists at any computable level, the identity G2 <= N + \\|B_N\\| is proven but degenerates to G2 <= N on this grid, and the sifting onset is flat at 1.63 to 1.70 against a sieve bound f2 that is zero everywhere the grid reaches. | 0 | [gap-spectrum-01.md](history/staging/gap-spectrum-01.md) |\n| `Q-gate-multiplies` | ANSWERED | Does \"the gate multiplies\" close the entire u-frame recursion branch, or only the merge chain? | Only the merge chain and one relative of it: the no-fixed-point argument does not reach TODO 0b at all, though 0b is wrong as stated for an unrelated reason and the error is a factor of ln u; what survives is the copy theorem for maxsum (VERIFIED 40 of 40), which needs a residue-deleted maxsum bound that does not pass through a kill count. | 0b | [gate-multiplies.md](gate-multiplies.md) |\n| `Q-gate-repair` | ANSWERED | Were the verify-the-verifier repairs landed with the gate ending fully green? | COMPLETE: node research/qc.js --full reads FULL GATE PASSED at the close, TOTAL 0 across the 11 checks, selftest 39 known positives firing and 32 controls silent, audit-numbers 246/246, and no bound tail left amber. | none | [gate-repair-finale.md](history/staging/gate-repair-finale.md) |\n| `Q-global-cutoff-averaging` | PARTIAL | Does the wider corpus suggest changing the global decomposition before extending the small-cofactor estimate, and can that change be made without an unpaid transition or outside term? | Derived: independently averaging the two initial Mobius cutoffs over fixed exponent intervals gives S=C2*x+sum G_L(n)G_R(n-2)+O_A(x/log^A x), uniformly on dyadic scales, with all cofactor and prime-power branches included. Each full G has squared norm O(x log x). The existing sharp-corner lower norm then forces cancellation between the corner coefficient and the rest at the same input. This does not estimate their shifted product. The global signed residual was already O(x) by the sieve upper bound and positivity; the new norm representation is not an improved signed bound. Prefer a bounded attempt on this global coefficient pair and a one-sided consumer; the twin margin remains OPEN. | C | [global-cutoff-averaging.md](global-cutoff-averaging.md) |\n| `Q-global-factor-signs` | PARTIAL | Which full factor configurations cause negative global products, can prime-power exceptions be paid, and does a pair-trigger upper bound control the negative part? | Derived: G_i(n)=Lambda_(>W_i)(n)-F_i(s_i(n))*log t_i(n)-E_i(n), with all small-prime factors in s_i and all primes of t_i exceeding W_i. The correction is supported on a small-prime power exceeding W_i; its full shifted effect is O_epsilon(x^(39/40+epsilon)). After also paying proper prime powers, the residual is a composite-filtered signed smooth/rough cofactor sum. A ten-small-prime cell has F=-84 although every pair lies below a, refuting the proposed pair-trigger majorant. Prime filters contribute explicit terms when bounding the negative part; dropping them is an upper bound, not an identity. No scale-x signed improvement or twin margin follows. | C | [global-factor-signs.md](global-factor-signs.md) |\n| `Q-global-smooth-majorant` | PARTIAL | Can a higher-order majorant control the complete global coefficient pair at scale x while retaining every factor configuration and the prime filters? | Derived using the corrected Henriot upper theorem: a C3 probability average of the same admissible initial cutoffs gives a full residual with sum \\|Ghat_L(n) Ghat_R(n-2)\\|=O(x). The three-smallest-prime majorant has bounded harmonic mass and meets the source growth class uniformly, despite not being multiplicative. All prime-power exceptions are paid. This replaces the O(x log x) absolute budget for the earlier logarithmic profile by O(x) for a different admissible profile representing the same signed residual to arbitrary logarithmic precision. The implied constant is not compared with C2 and no improved signed lower bound or twin margin is supplied. | C | [global-smooth-majorant.md](global-smooth-majorant.md) |\n| `Q-greedy-oracle` | ANSWERED | Is the corrected greedy a G2 oracle? | ORACLE ESTABLISHED, 13 of 13 exactly-known terms hit with minimum ratio 1.0000, so the greedy RULE is an exact solver where the answer is checkable at x <= 41; the rider matters more, the search budget needed grows 3.31x per additional prime and two objects whose truth reaches further show the same estimator's fidelity DECAYING, so nothing is established for the greedy AS RUN on the ladder. | 1b (retired) | [greedy-oracle-validation.md](history/staging/greedy-oracle-validation.md), [phase1-T1-greedy-oracle.md](history/staging/phase1-T1-greedy-oracle.md) |\n| `Q-grouped-divisor-moment` | ANSWERED | Does the full gcd-normalized moment proposed by the literature audit hold, and what exact portion of the twin-prime remainder does it control? | The proposed moment is derived from classical completion with all coefficient sectors and uniform twists included. Full rectangles are controlled when delta<19/25 and delta+3nu<161/100, in addition to the preceding region. A concrete extra cut d<=floor(x^(151/200)), de^3<=floor(x^(321/200)) controls the entire d~e~x^(2/5) benchmark and leaves an explicit smaller-domain endpoint remainder. The uniform product threshold stays below 19/25. At delta=8/25,nu=9/20, the next deficit is confined to small-common-divisor nonzero kernels; their required saving and the global twin margin remain OPEN. Finite validation does not prove asymptotic rates. | C | [grouped-divisor-moment.md](grouped-divisor-moment.md) |\n| `Q-growth-law` | ANSWERED | What does G2(x#) actually grow like? | MEASURED about 0.762 * x ln^2 x * lnln x over x = 11..79 (18 terms, ten log-linear families AICc-ranked), with the honest statement being the band and not the winner; the pure power law is excluded against a control that knows its own answer, and G2/h costs one further logarithm. | none | [attack-growth-law.md](history/staging/attack-growth-law.md) |\n| `Q-h2-extension` | CLOSED | Is extending A288815 (h2) past its 21 terms the high-value computation it was taken for? | No, on both halves, and it is recorded in OUTCOMES.md as WITHDRAWN: term 22 is a roughly four-year single-thread computation for the specialists' own tuned code, all three cheaper proxies fail downward and growing (the direction that fakes a smaller exponent), and the control measured one extra term moving a ten-term exponent fit by 0.022. | none | [h2-scoping.md](h2-scoping.md) |\n| `Q-handoff-review-0906` | ANSWERED | Does the handed-back arithmetic campaign survive an independent check of its main regional estimate and the conclusions used to choose the next research direction? | The bounded handoff audit is completed in reports 20 and 21: the checked local reduction and regional mechanisms survive, named source statements were verified, and the consumer, corner support, rate, shrinking-margin and identity-piece overclaims were corrected. Joint Cauchy is now priced and adds no region. The multiplicative band transfer has a separate PARTIAL owner with a weaker continuous-scale payoff. Imported deep theorems remain imports; this is not corpus-wide certification or a twin margin. | C | [handoff-review-0906.md](handoff-review-0906.md) |\n| `Q-hawkins-read` | ANSWERED | Does the Hawkins random-sieve tradition, read at source, hold a two-class variant, a stage-to-stage moment identity, or an answer to the analogue question? | NOT FOUND for a two-class or k-residue variant; the moment identity is found in the null and not in the fold; and the tradition raises the analogue question repeatedly and answers it only as a missing constant. Two primaries stayed paywalled, so the absence claims rest on the rest of the bibliography. | none | [hawkins-read.md](history/staging/hawkins-read.md) |\n| `Q-head-remainder-0830` | PARTIAL | Does the 5.3 percent remainder that Hardy-Littlewood leaves in the head's endpoint deficit Delta = 2 - beta derive by a route that is not HL: the frozen-sieve first-survivor object, a level-of-distribution correction, or the exact tile-ensemble head? | No route that is not HL closes it, and the remainder as recorded is not what it was taken to be. The record's Delta_meas = 0.6214 at [1e7,1e8) pools a decade over which the prime density falls ten percent, and the OLS intercept absorbs the gradient (a Simpson-type effect, sign derived): the two half-decades read 0.6848 and 0.6500, the four quarter-decades 0.7203 to 0.6565, and pooled minus the gap-weighted mean of the halves is -0.0378, -0.0455, -0.0524 at the three decades, so the same-sign 5.3 percent miss is the pooling. At half-decade resolution HL's pipeline prices Delta to 1.2 and 0.4 percent at the top two windows, inside one standard error, and the sign of the miss alternates. That agreement is a near-cancellation of two four-percent effects: the pipeline under-predicts its own rotation ensemble by 3.6 to 4.9 percent (pipe/ens 0.951 to 0.964 across y = 313 to 19997, MEASURED by exact CRT sampling; 1.049 to 0.958 at y = 11 to 23 by full enumeration), an ensemble quantity that derives by finite count with no prime input and whose sign is opposite to hl3 section 2's guess; and the anchored window's deficit sits 2 to 5 percent below its gap-scale-matched ensemble (Delta_meas/Delta_ens 0.878 to 0.995 over six half-decades, 0.954 to 0.995 on the four with more than 13,000 gaps), the anchoring part, which is a three-point prime correlation in a short window and is HL-strength by every route named. The head is a slack field and nothing here touches Z2. | Z4 | [attack-0830-head-remainder.md](history/staging/attack-0830-head-remainder.md) |\n| `Q-head-residual` | PARTIAL | What is the head's residual prime-origin factor h/R = 1.09 -> 1.03, and does it derive? | h - R decomposes exactly and A_forced derives; h/R -> 1 follows from beta CV^2 being bounded, beta CV^2 -> 2 controlling the RATE and not the limit (under it h - R -> D = -0.95, so +8.4661/ln^2 p cannot be the limit law); R is the continuum functional and the integer-origin null sees R + 1/2 exactly, so h - R = 5.679 depends on the unstated null population, reading 5.179 discrete-uniform and 2.925 coprime-to-30; the mod-30 answer is NO by a valid route, class term -0.0001; Delta = 2 - beta is priced by HL to 5.3 percent with nothing fitted and HL's own 1/ln x term absorbs the miss, so the 5.5 s.e. framing is dropped; the asymmetry-as-mechanism claim and the [0.289, 0.361] bracket are refuted, hl3's 16 percent standing; beta_null is 1.898 for the discrete-consistent null and 2 in the continuum; the 72/28 split reads 81/19 in the other exact order; c = 0.844 is 0.722 on a subset, pinned to 15 percent; Delta_HL ln p/lnln p = 4.02 is a property of the prediction; and the N4 plateau is not OPEN, needing beta CV^2 = 2.71 rising to 6.63. | Z4 | [head-residual-factor.md](history/staging/head-residual-factor.md), [head-residual-hl3.md](history/staging/head-residual-hl3.md), [head-residual-null.md](history/staging/head-residual-null.md) |\n| `Q-heath-brown-edges` | ANSWERED | Does the a=1 edge, and hence the corner S_0, come from the particular two-cutoff Vaughan identity in prime-detection-spec.md and shifted-prime-decomposition.md, or does an equivalent edge reappear under Heath-Brown's identity and the other standard decompositions of Lambda? | The exact identities and exponent simplex leave a=1 in separately bounded formal pieces, including Heath-Brown's balanced j=K configuration. This does not prove an intrinsic edge after identity-piece cancellation. The bounded-cofactor object has affine dilations, not literally F(n)F(n-2). Linnik's fixed truncation has a large term-wise absolute tail by an elementary counting proof; its signed tail is not thereby bounded below. Heath-Brown K=3 narrows the classical split band, with coefficient and Type I normalization obligations retained. No full alternative reduction, new region or twin margin follows. | C | [heath-brown-edges.md](heath-brown-edges.md) |\n| `Q-history-dial` | ANSWERED | Does revealing the scour primes' classes concentrate the floor on a bounded set of strike locations, and what is the history dial worth? | Disconfirming: the class axis is linear at 2.90 survivors per prime read, so no bounded set of strike locations carries the floor, and the touch-count ceiling is proven and tight; deliverable (1) of the brief already existed as canonical state (the forcing ladder 16 -> 45 at @11), so nothing new was bought there. | A | [attack-history-dial.md](history/staging/attack-history-dial.md) |\n| `Q-hm-basis` | ANSWERED | Does the u_sup divergence survive the change to the natural (h,m) basis? | It survives: the two bases are nested rather than rivals, the natural one is the outer and weaker member, and the gap between them is a flat factor of 7.26 across nine levels, so the divergence that closed u_sup runs in both bases. | none | [attack-hm-basis.md](history/staging/attack-hm-basis.md) |\n| `Q-holt-2605` | ANSWERED | Does the fifteenth Holt manuscript, arXiv:2605.19165, contain anything that touches our record? | No: it carries no twin, no max-gap bound and no G2, its \\|s\\|/2 threshold restates the 2603.25915 / 1408.6002 inequality, and the interaction check is negative, our extinction law diverging from his by a factor 396 across three decades of W; the sweep is now complete at fifteen of fifteen. | none | [holt-2605-sweep.md](history/staging/holt-2605-sweep.md) |\n| `Q-hsub-reductions` | PARTIAL | Can (H-sub) be proven or refuted, and what does the fold machinery reach? | Neither, but the statement is smaller: three reductions of the hypothesis are proven, the lemma consuming only power pairs so that it weakens to (H-sub-pow) with the conclusion unchanged, the fold machinery's inability to reach any of them is made exact, and the hunt over all 111 reachable pairs found no drifting family and no counterexample. | 1d | [attack-hsub-01.md](history/staging/attack-hsub-01.md) |\n| `Q-hsubpow-K` | CLOSED | Can (H-sub-pow) be proven with an explicit K in the legal zone? | NO by three mechanisms: the CRT lift (K* >= pi(y') - pi(y) diverges), anchored caps (diverge at every base), Iwaniec at two classes (it IS beta2-note); TODO 1d's legal zone is mis-stated, the trusted zone is [1.3946, 11.3568). | 1d | [hsubpow-explicit-K.md](history/staging/hsubpow-explicit-K.md) |\n| `Q-hsubpow-K-0829n` | OPEN | Can (H-sub-pow) be proven with an explicit K inside the trusted legal zone [1.3946, 11.3568) by a mechanism the 2026-08-28 pass did not close? | No K is proven at any base; the single open inequality is the uniform-in-k ratio cap G(b^(k+1))/G(b^k) <= e^K G(b), which is a proof gap at a fixed base and a possible truth gap across bases, since for any law G ~ c n^beta (ln n)^delta the all-bases hypothesis holds with finite K if and only if delta >= 0. | 1d | [attack-0829n-hsubpow-K.md](history/staging/attack-0829n-hsubpow-K.md) |\n| `Q-hybrid-bound` | CLOSED | Does an exactly-handled head of small primes plus a sieved tail improve the certified exponent? | The glue closes and is a real theorem, and it buys one factor of 1.93 in x and nothing asymptotic: the window where the bound beats 4.2665 widens from x <= 227 to x <= 439 and never returns, the head can only reach x0 = O(ln x), and the literal glue the brief named (A144311 as base case) is dead on a type mismatch. | none | [attack-hybrid-bound.md](history/staging/attack-hybrid-bound.md) |\n| `Q-identifications-prior-art` | ANSWERED | Who owns the three identifications that were banked without a prior-art search? | All three are OWNED in their conventions: the complementary-window duality, maxsum_m as the largest m-spacing of a scan statistic, and the alternation language as a textbook charge constraint; the exponent law H = a + b ln D and the H* Shearer-threshold ladder come back NOVEL-SO-FAR on four calibrated channels. | none | [identifications-prior-art.md](history/staging/identifications-prior-art.md) |\n| `Q-import-bfree` | ANSWERED | Does the ergodic theory of B-free systems - Mirsky measures, Toeplitz structure, tautness, heredity, entropy - say anything about the twin comb? | It lands as an exclusion: the comb is exactly the object the field's newest generalisation is built for and that generalisation names the comb's own regime as the one it declines to study, the limit comb is the single point {-1} so every ergodic statement about it is vacuous, and the yield is a corrected address for the walls plus one live external open problem. | none | [import-bfree.md](history/staging/import-bfree.md) |\n| `Q-import-boolean-analysis` | CLOSED | Does the analysis of Boolean functions reach the anchored bias? | Five things that had to hold do not, starting with the product space eating the statement worse than C^pi(x); the forced degree saturates at 2 and the globality hypothesis fails at q1, so the row supplies a wall address rather than a route, and every figure is SCRATCHPAD-GRADE and may not be quoted until re-derived inside an embedded producer. | none | [import-boolean-analysis.md](history/staging/import-boolean-analysis.md) |\n| `Q-import-bridge` | CLOSED | Do Ojaroudi's Bridge and collision lemmas certify a number the corpus lacks? | Both lemmas are true and both proofs go through in three lines here, and the answer to the question the import was raised for is no: the certified-number verdict is zero by an identity rather than by a margin, and the per-prime nature is what stops it. | none | [import-bridge.md](history/staging/import-bridge.md) |\n| `Q-import-bridge-prereg` | SUPERSEDED | Do Ojaroudi's two elementary lemmas, imported onto our tile, beat the incumbent per-prime discrepancy numbers? | Pre-registration only, sealed before any number was computed: the predictions, the incumbents, the win condition and five declared traps are fixed here, and the null result is the expected result because P4 is an identity; the outturn is import-bridge.md. | none | [import-bridge-prereg.md](history/staging/import-bridge-prereg.md) |\n| `Q-import-chaining` | CLOSED | Can Dudley's entropy bound or Talagrand's generic chaining beat the union bound over positions on the maxsum law? | No, and the reason is geometric rather than probabilistic: the entropy integral of the true increment metric is already 1.054-1.099 times rms*sqrt(2 lnW) at its lower branch and 1.484-1.545 at its upper, flat across z = 13..23, so chaining's ceiling with a perfect universal constant sits below the union bound's floor; the honest constant-carrying chain measures 2.878 to 3.027. | 0c | [import-chaining.md](history/staging/import-chaining.md) |\n| `Q-import-distortion` | CLOSED | Does the distortion method of covering systems speak at two classes per modulus, and what does it certify on the window? | It does speak at two classes, and the row's expected blocker was the wrong one: its economy sum 4/p^2 converges (sup 0.364545) where the union bound's sum 2/p dies at x = 13, but it charges in ambient length, its measures living on a CRT product so the certified window is at least a primorial and the only published bridge back to an interval is exponential in the number of progressions. | none | [import-distortion.md](history/staging/import-distortion.md) |\n| `Q-import-distortion-prereg` | ANSWERED | What does the distortion method certify on our window, and does it speak at one class per modulus or at two? | Sealed before any line of the producer existed; scored in import-distortion.md, where the deciding question comes back opposite to the row's expectation: BBMST's criterion carries no distinctness, multiplicity or minimum-modulus hypothesis, so the method speaks at two classes and the one-versus-two-class wall is not its wall. | none | [import-distortion-prereg.md](history/staging/import-distortion-prereg.md) |\n| `Q-import-entropy-decrement` | CLOSED | Does Tao's entropy decrement argument, the one place a parity-type obstruction was circumvented for a two-point correlation, transfer to the tile's census? | It does not: the engine runs on multiplicativity at small primes and the census is defined by the absence of small prime factors, Tao scopes twin-prime sums out himself in print, and the census sits on the structured side of the pretentious dichotomy; all figures SCRATCHPAD-GRADE. | none | [import-entropy-decrement.md](history/staging/import-entropy-decrement.md) |\n| `Q-import-fracparts` | ANSWERED | Does Saffari-Vaughan Theorem 10 cover the branches carrying the Skeleton door's mass? | Prediction HIT: the covered branches carry at most 8 percent of the mass; anchor and a sharper wall-address banked; the THEOREM column is struck, not merely overpriced, because Theorem 10's range condition is x^{6/11+eps} < y, which fails at fixed M; no route; no cap-36 line is upgraded. | 4 (retired) | [import-fracparts.md](history/staging/import-fracparts.md) |\n| `Q-import-hypergraph` | MIXED (import-hypergraph-prereg.md: OPEN; import-hypergraph.md: ANSWERED) | Does hypergraph covering (Pippenger-Spencer / FGKMT) apply to the adversary's construction? | The covering theorem is class-count-agnostic and the analogy is EXACT at the hypothesis level, leaking at exactly one named place, the edges the machine needs as input; Branch A landed a theorem-provenance lower bound G2(x#) >> x ln x (HELD), Branch B died as predicted with a sharper wall than the registered one, and none of the three pre-registered kill lines fired. | none | [import-hypergraph-prereg.md](history/staging/import-hypergraph-prereg.md), [import-hypergraph.md](history/staging/import-hypergraph.md) |\n| `Q-import-interp` | ANSWERED | Does the Bayati-Gamarnik-Tetali interpolation method reach the G2 Fekete object? | No: BGT closes on hypothesis H1, since the constraint count never splits (pi(st) - pi(s) - pi(t) is zero at no pair with st >= 25, minimum 2, maximum 12, VERIFIED on 104 pairs), and the one coordinate that does split has limit +infinity; the row lands as WALL-ADDRESS plus the identity that the submultiplicativity defect is the Overshoot slack. | 1d | [import-interp.md](history/staging/import-interp.md) |\n| `Q-import-interp-prereg` | SUPERSEDED | Does the interpolation method (Bayati-Gamarnik-Tetali, Proposition 5) transfer to the G2 ladder, and in which coordinate? | Pre-registration only, committed alone before any producer existed, with the source verified at page images first (Proposition 5 is a special case of de Bruijn-Erdos, by the authors' own attribution); the predictions and the meaning of each outcome for the map are fixed here, and the outturn is import-interp.md. | none | [import-interp-prereg.md](history/staging/import-interp-prereg.md) |\n| `Q-import-kw-zonegap` | ANSWERED | What does the graded Kourbatov-Wolf row buy the Z2 programme, and does the flagged 0.7574 against 0.7504 agreement corroborate anything? | A split-cell circularity verdict: the trend formula's raw performance is real, G/Tbar inside [0.78, 1.05] over thirteen decades with zero fitted parameters, but the flagged agreement is adjudicated a range accident, and since Theorem occurs zero times in the source the row can only be priced PUBLISHED-ANCHOR plus WALL-ADDRESS, whose wall is a quantifier: an almost-all over a log-sparse record sequence cannot deliver an every-zone statement. | none | [import-kw-zonegap.md](history/staging/import-kw-zonegap.md) |\n| `Q-import-l1l2` | CLOSED | Does an l1 -> l2 conversion on the (e,a) coefficient sum buy the needed sqrt(log)? | The pre-registered kill line (rho >= 0.5 at every level) does NOT fire, rho falling 0.4352 to 0.0645 at z = 13..41, and the programme dies anyway on what the decay measures: \\|\\|Theta\\|\\|_1/\\|\\|Theta\\|\\|_2 grows by 2.01 per added prime against 2.0516 for S_sat, so the l1/l2 gap is not a proxy for the C^{pi(z)} price, it IS that price; rows 5 and 6 retire together. | none | [import-l1l2.md](history/staging/import-l1l2.md) |\n| `Q-import-l1l2-prereg` | CLOSED | Does the l1 to l2 conversion on Theta_e(a) reach the threshold the import map set? | Sealed before any number was computed; scored in import-l1l2.md, where the pre-registered kill does not fire and the programme dies anyway: rho falls monotonically from 0.4352 to 0.0645 over z = 13..41, never reaching the map's 0.5 line, and what the conversion outputs is the sharp maximal law. | none | [import-l1l2-prereg.md](history/staging/import-l1l2-prereg.md) |\n| `Q-import-map` | PARTIAL | Which foreign mathematics is worth aiming at this object, and what did each landed import actually buy? | Nineteen landings, zero routes opened on the exponent; every one banked a proven identity, a derived constant, a published anchor or a wall address, which is why payoff is graded by TYPE rather than by probability, and the map is live with rows still unrun. | none | [IMPORT-MAP.md](IMPORT-MAP.md) |\n| `Q-import-map-construction` | ANSWERED | What was searched, what was rejected, and at what verification level, in building IMPORT-MAP.md? | The reconnaissance record behind the map: candidates rejected with reasons (hypergraph containers for want of a supersaturation input; determinantal processes by a proof-level obstruction, the twin tile failing the pair inequality at d = 6), a per-attribution SOURCED / SOURCED-BIB / MEMORY ledger, and two results derived here and offered for adjudication. | none | [import-map-construction.md](history/staging/import-map-construction.md) |\n| `Q-import-maxplus` | ANSWERED | Do max-plus (tropical) spectral theory and subadditive ergodic theory bound the fold recursion's growth exponent? | The mapping is exact rather than an analogy - the fold is p-fold cyclic duplication followed by max-plus state elimination and the copy theorem is a semiring identity, verified at 6 of 6 folds - but the tropical Perron root of the tile is the mean gap mbar, not G2, so the import delivers a structural negative, a sharpening of A5's Theorem A, and refuses the growth exponent. | none | [import-maxplus.md](history/staging/import-maxplus.md) |\n| `Q-import-proof-complexity` | CLOSED | Does proof complexity supply the barrier that wall-note Face 4 wants? | The row is VOCABULARY-ONLY and is recommended for rejection at the gate: the shared word is parity naming two different objects, a degree bound is a pseudo-expectation whose finite exact instance the corpus already refuted, and any unconditional barrier would imply Siegel-zero finiteness; the payoff is Tao and Brady 2014 read at source as a published anchor and a wall address, not a theorem. | none | [import-proof-complexity.md](history/staging/import-proof-complexity.md) |\n| `Q-import-repulsive` | CLOSED | Does any repulsive point-process model (determinantal, Coulomb gas, hyperuniform) reproduce the twin tile's sub-Poisson counts? | NO, two-sided: g(2) = 0 kills positively-associated models and g(6) = 2.66 (g never in (0,1); support 6 \\| d, g >= 2.38) kills Hermitian determinantal and negatively-associated ones; the Gibbs extension is withdrawn; every fixed level is periodic hence class I hyperuniform and the statement is empty, while along the diagonal family L = y^u no member approaches a hyperuniform limit in Var/E, which is a statement about the family and not a class for any one tile; the determinantal row's quoted product evaluates to 0 and its obstruction distance is 6 not 2. | none | [import-repulsive.md](history/staging/import-repulsive.md) |\n| `Q-import-rough-anatomy` | ANSWERED | Does the analytic theory of rough numbers derive the depth law's drift, and is u = 2 a singular point of the dimension-2 sifting functions? | SCRATCHPAD-GRADE throughout, and leaning on three never-red-teamed premises: the drift is derived to 0.16 per cent by exact Mertens products and NOT by any published next-order term (the one in print is the wrong term for this object), and u = 2 is one identity seen four times rather than a singular point of the dimension-2 sifting functions. | none | [import-rough-anatomy.md](history/staging/import-rough-anatomy.md) |\n| `Q-import-rows-15-17` | ANSWERED | Which three new fields deserve import-map rows, and at what price? | Row 15 is a wall address and a published anchor and NOT a theorem, since Theorem 10 requires x^{6/11+eps} < y and fails at every fixed M; row 16 is stale, its experiment already executed by record-location-null.md and its prior-art half closed by lit-kourbatov-shortfall.md; row 17 lands as priced; six fields rejected with pointers and one promoted to row 17. | none | [import-map-rows-15-17.md](history/staging/import-map-rows-15-17.md) |\n| `Q-import-scanstat` | ANSWERED | Does the extreme-value theory of scan statistics settle the disputed sqrt(m) claim of IMPORT-MAP row 1? | The identity offered for adjudication is true and its conclusion is false - Sum gamma(k) = 0 is trivial while the step to Var(S_m)/m tending to 0 is refuted by a free permutation counterexample - the Leadbetter half of the row is confirmed as written and the pre-registered kill criterion passes at both target levels, one of them blind; the adversary pass corrected the claim that T29 had never been computed here to a reproduction. | none | [import-scanstat.md](history/staging/import-scanstat.md) |\n| `Q-import-scanstat-prereg` | OPEN | What is the maxsum exponent at T_23 and T_29, predicted from a T_13/T_17/T_19 fit before the run? | Machine-written pre-registration only: the extrapolation laws, the grid and the kill criterion (Model A must beat Model B on the ln-RMS of excess_m at T_23) are fixed before any scoring, so part 4 cannot be scored against a prediction chosen after the fact. | none | [import-scanstat-prereg.md](history/staging/import-scanstat-prereg.md) |\n| `Q-import-shearer` | ANSWERED | Where does Shearer's region actually end for the fold read as a repulsive lattice gas? | SURVIVES WITH CORRECTIONS under the 2026-08-19 adversarial pass: the complete-graph identity is verified by independent-set enumeration at n = 2..8, H* = 35..481 reproduce exactly and the 2/sqrt(e) floor is sound, but the tightness witness's disjoint-intervals sentence is false exactly where invoked and the atomicity lemma's hypothesis is load-bearing and not shown, the closure surviving instead on the sequel's gamma_i >= mu(f_i). | none | [import-shearer.md](history/staging/import-shearer.md) |\n| `Q-import-shearer-prereg` | ANSWERED | Where does Shearer's region actually end on the repulsive lattice gas model of our window? | Sealed before any line of the producer existed; scored in import-shearer.md, where the exact criterion says the local lemma is the union bound on a complete dependency graph, and the two levels it buys over the local lemma's own sufficient condition are the only thing it buys. | none | [import-shearer-prereg.md](history/staging/import-shearer-prereg.md) |\n| `Q-import-sofic` | ANSWERED | Does constrained coding and symbolic dynamics identify the Alternation Lemma's shift, and does the import map's shape claim survive? | The identification is exact and better than the map states - the shift is strictly sofic, two states, four edges, Perron root exactly 2, and the graph's own per-step rate is 2/p rather than the 3/p the map divides by - and the pre-registered ratio test then kills the map's literal formula while the graph-corrected one survives only as a small-D transient spent by fold 37; the shape claim does not stand, and it fails in the corpus's favour. | none | [import-sofic.md](history/staging/import-sofic.md) |\n| `Q-import-sofic-prereg` | OPEN | Does the constrained-coding presentation of the fold predict the exact per-fold L, and is the shift sofic? | Pre-registration only, and declared weaker than the G2(41#) case because the nine exact L values were already published in the corpus: what is fixed here is the estimator, the exclusion rule, the trend statistic and the kill thresholds, plus a soficity verdict written before the check, strictly sofic and not of finite type. | none | [import-sofic-prereg.md](history/staging/import-sofic-prereg.md) |\n| `Q-import-stein` | CLOSED | Does Stein's method for Poisson approximation derive the extinction law's constants and the ~3.8 joint deficit? | The pre-registered kill fires at both targets for one structural reason: each object is driven by exactly one shared uniform draw, so the non-neighbourhood indicators reconstruct it and b3 climbs to its own ceiling, measured at 0.850 to 0.9998 of that ceiling and 200 to 800 times b1 + b2; with no error term there is no Stein derivation, and the first-moment arithmetic banks the (A, c) pair anyway, relabelled. | none | [import-stein.md](history/staging/import-stein.md) |\n| `Q-import-stein-prereg` | CLOSED | Does Stein's method for Poisson approximation bound the extinction law or the ~3.8 constant? | Sealed before any producer ran; scored in import-stein.md, where the pre-registered kill criterion fires at both targets for one structural reason, that both objects are driven by exactly one shared uniform draw, which is precisely the dependence the b3 term measures. | none | [import-stein-prereg.md](history/staging/import-stein-prereg.md) |\n| `Q-import-suen` | ANSWERED | Do Suen's inequality, Janson's inequality and the Lovász Local Lemma family supply the missing proof for the anchored delta? | No inequality can: the import lands and what it certifies is that the target was never a lemma, the NOT-REACHED item being the Twin Prime Conjecture wearing a percentage; what it does deliver is the constant in the sharp form and the wall's address in local-lemma coordinates. | none | [import-suen.md](history/staging/import-suen.md) |\n| `Q-import-thinning` | ANSWERED | Do point-process thinning and interval coalescence give H'' a null model? | Yes and exactly: the fold recursion is an exactly solvable thinning whose gap pgf transforms by a Mobius map fixing 0 and 1, so under independent thinning the gap word of T_x is exactly geometric with mean mbar/6 at every rung with no error term, which turns both of the record's fitted parameters into derived ones. | 0b | [import-thinning.md](history/staging/import-thinning.md) |\n| `Q-import-transference` | CLOSED | Does the transference principle and relative Szemeredi theory reach the twin pattern? | VOCABULARY-ONLY and closed by an exclusion in print: the founding paper names our configuration as out of scope in one printed sentence, the theorem degenerates at k = 2 which is the case we want, and the W-trick removes the tile before the argument starts; re-entry costs a finite-complexity system whose solvability implies the Zone Postulate. | none | [import-transference.md](history/staging/import-transference.md) |\n| `Q-import-vc-nets` | CLOSED | Does VC dimension, epsilon-net or set-cover duality bound G2? | VOCABULARY-ONLY, at a self-set 0.2 per cent confidence of opening the 4.2665 -> 2 gap and with every number SCRATCHPAD-GRADE: the range space's VC dimension is small and its growth is O(log log L) (proven), but the dual shatter function is exponential, the interval formulation's hypothesis contains its conclusion, and the full-period formulation is clean, free and on the wrong side of the inequality. | none | [import-vc-nets.md](history/staging/import-vc-nets.md) |\n| `Q-ioslack-survey` | CLOSED | Does any instrument in the corpus carry a sharp-level signature that would let the infinitely-often slack be spent? | None does: 23 instruments surveyed and 19 eligible, smallest per-instrument p_FWE = 0.1082 and 0.8865 after Šidák, so dial 4's fourth channel closes and the dial is finished; the second finding is that maxsum_{L+1} is sharp exactly when L = 1. | 0e (retired) | [ioslack-survey.md](history/staging/ioslack-survey.md) |\n| `Q-joint-correction-source-audit` | ANSWERED | Does completing the supported prime-partner correction yield cancellation beyond the existing Vaughan reduction, and which literature inputs actually justify that completion? | CHECKED prior-art matches for subset convolution, complementary sieve sums and signed smooth-number distribution. DERIVED reconstruction identifies the unrestricted factor weight with an existing short divisor approximant, up to the paid prime-power error. Its supports are x^(12/25) and x^(1/10). Reassembling the entire joint correction returns the already-open Vaughan remainder and supplies no new signed bound. A formal factor-vector control distinguishes the unrestricted logarithmic weight from the selected prime-cofactor branch. The matched general sieve theorems require inputs not supplied here. Literature novelty is unestablished; twin infinitude remains OPEN. | C | [joint-correction-source-audit.md](joint-correction-source-audit.md) |\n| `Q-joint-factor-estimate` | ANSWERED | Can an additional estimate for actual factor configurations control the full signed balance of the C3 residual? | DERIVED: for the single-large-prime delta-rough family F, the residual splits exactly as R_F = C2*x*H_delta - Z_F^p + o(x), where the Type I part C2*x*H_delta is evaluated by ordinary prime BV on each cofactor interval and the twisted Mobius constant 2C2 (corrected 2026-09-08 after V2: Wu Lemma 2.3 is squarefree-only and did not cover the non-squarefree b_R moduli; prime BV covers all moduli), and Z_F^p is the signed prime-partner sum. The complement carries the opposite Type I part -C2*x*H_delta, so grouping by factor configuration transfers Type I mass and leaves the twisted prime correlation B_L unchanged. With the matched upper sieve the family bound is C2*x(1+H_delta-(40/9)H^+), certified below -2/5 C2*x: INSUFFICIENT BOUND. Even a sieve constant 2 in place of 40/9 needs H^abs<1 and leaves the complement unpaid. Independently reviewed 2026-09-09 with the uniform delta-range and complement corrections below. Twin infinitude and every sufficient margin remain OPEN. | C | [joint-factor-estimate.md](joint-factor-estimate.md) |\n| `Q-k4direct-swap` | ANSWERED | Does swapping the shipped Monte Carlo estimator for k4direct meet the @13 gate, and is the x3600 on record a speedup? | The swap meets the gate by x3632, but the margin on record is an ACCURACY margin and not a speedup: k4direct is a 4x to 8x SLOWDOWN, 93.5 s against 11.7 to 23.4 s, and the 198.9 s on record for it is 2.1x too expensive, which drags the @17 extrapolation from 8.5 days on ten cores to about 4.4. | none | [verify-k4direct-swap.md](history/staging/verify-k4direct-swap.md) |\n| `Q-kappa-not-L` | CLOSED | Can L, or its generalisation kappa(m), be bounded well enough to be the route's target? | L is not the target: three bounds on L are PROVEN through the Alternation Lemma and transfer to kappa(m), but the wall is located precisely and the lower bound never becomes exact, so L cannot be made irrelevant; closed 2026-08-19, L has no law of its own, being G2 divided by the block's own spacing and super-additive on disjoint sets. | none | [kappa-not-L.md](kappa-not-L.md) |\n| `Q-kernel-sign-control` | ANSWERED | At delta=8/25, nu=9/20, is the moment Mfrak and its small-common-divisor kernel X_small measurably smaller with the actual coefficient b_u=A_right(gu) than with \\|b_u\\| or with \\|b_u\\| times an independent seeded random sign, and does the actual/random ratio decrease with x over the feasible dyadic range? | MEASURED, negative. Over seven dyadic scales j=18..30 and four boxes, the pre-registered slope of log2(actual/random) for X_small has mean + sd >= 0 in all four families, so the run gives no heuristic support for extra Mobius-sign cancellation in the small-j kernel. The actual coefficient's \\|X_small\\| ranks like one more random draw. The measurement is weak where it matters: the finite moment is 97.66-102.58 percent equal-frequency class, X_small carries 0.02-2.88 percent, J0=floor(x^(1/20)) is 1 at j=18 and 2 at j=20..30, and Z=2 collapses the prime-power sector of (13) to r=2. Equation (21) and the global twin margin remain OPEN and untouched. | C | [kernel-sign-control.md](kernel-sign-control.md) |\n| `Q-kk-substitution` | ANSWERED | Does the Kalmynin-Konyagin construction survive substituting the two-class set Omega_p = {a_p, a_p - 2}, and what lower bound does it give? | STANDS WITH CORRECTIONS: G2(P(y)) >> y (ln y)^3 (lnlnln y)^2 / (lnln y)^4 for y >= y0 survives every attack made here, including an exhaustive brute force of the Proposition itself over four million values of i; five corrections follow and none touches the exponent. | none | [attack-kk-substitution.md](history/staging/attack-kk-substitution.md), [verify-kk-substitution.md](history/staging/verify-kk-substitution.md) |\n| `Q-klz-forward-walk` | ANSWERED | Does the forward citation graph of arXiv:2205.08273 hold a two-class or k-class complexity result, and is Part II out? | The forward graph has exactly one member on three independent indices and it never treats word complexity; no citing work goes beyond one class per prime, Part II is NOT OUT as of 2026-08-19, the owning convention was OEIS A023192 all along, and the comparison is APPLES-TO-APPLES so the exponent-shape sentence needs no heredity caveat. | none | [klz-forward-walk.md](history/staging/klz-forward-walk.md) |\n| `Q-kstar-drift` | ANSWERED | Is the drift in K* across doubling steps structural, and how far past the scannable levels does the certificate reach? | Structural and it lands early: K* is now enumerated at 16 doubling steps against 11, the raw drift slope steepens from 0.6881 +/- 0.1328 to 0.8184 +/- 0.0908, and the period-free certificate touches K*+1 = 18, 93.5% of 2^beta2 = 19.2455, at the base-2 chain step s = 16 itself. | none | [attack-kstar-01.md](history/staging/attack-kstar-01.md) |\n| `Q-kstar-prereg` | OPEN | What is K* at the three next doubling steps, predicted before any period walk? | Pre-registration only, committed alone: the predictions, the scoring rule and the growth-type verdict thresholds are fixed in advance, with the inclusion-exclusion engine validated against an independent scan engine on all eleven known steps first. | D | [attack-kstar-01-prereg.md](history/staging/attack-kstar-01-prereg.md) |\n| `Q-l1-residue` | CLOSED | Is the L = 1 residue count an open counting hypothesis whose proof would deliver the Zone Postulate? | REFUTED as a hypothesis: restore the dropped L >= 2 terms and it IS the Zone Postulate, so the chain is a tautology, and leave them out and what remains is provably no stronger than the target; Lemma A (inside B(p) the residue condition is the kill condition) is PROVEN and VERIFIED at 237 folds, and the route lands back on the residue-deleted maxsum without advancing it. | none | [attack-l1-residue.md](history/staging/attack-l1-residue.md) |\n| `Q-lambda-ledger` | CLOSED | Does parity information survive in the lambda-weighted fold ledger? | REJECT, against a pre-registered threshold fixed before the producer existed: the lambda multiplier exists and is parity-free, so the ledger is blind at its design point rather than at its margins; two columns are pinned by proof, and nothing here is novel mathematics. | Z5b (retired) | [attack-lambda-ledger.md](history/staging/attack-lambda-ledger.md) |\n| `Q-ledger-extraction` | ANSWERED | Can the QC engine's three suppression ledgers be moved out of the checks and into data? | Moved: all three now live in research/qc/ledgers.js, the fast gate and the selftest are byte-identical to their pre-change output, and all 25 entries were verified verbatim against git HEAD by key, order and reason text. | none | [audit-ledger-extraction.md](history/staging/audit-ledger-extraction.md) |\n| `Q-ledger-forced` | ANSWERED | Which properties of the fold ledger's columns are FORCED rather than expected? | Fourteen forced constraints, all identities or wrong-direction bounds; by_new is forced only in the prime gap (its O(1) is MEASURED); nothing forces net from below. | Z5b (retired) | [fold-ledger-forced.md](history/staging/fold-ledger-forced.md) |\n| `Q-left-divisor-signs` | PARTIAL | Can the Möbius structure of the left coefficient A_left(gm) replace the first Cauchy inequality, so that its sqrt(M) loss is not paid, and what does that buy at the target box (delta,nu)=(8/25,9/20) and at the corner a=b=1? | Partly, and not where it is needed. Two lemmas are derived: a Type I bound with no zero-frequency budget, usable at the target box whenever the arbitrary block is shorter than x^(1/4), and a Type II bound that is uniform in the Perron twist heights. Applied to the actual coefficient they confine the target-box deficit to sectors with left prime power r>=x^(43/200) and right prime power q>x^(3/100), and lower the worst block exponent there from 103/100 to 41/40. The corner a=b=1 is untouched: grouped, Type I, Type II, Bettin-Chandee/Wright and even a hypothetical per-block square-root bound all exceed 1 there. No region is added, E_dagger and the consumer (20) are unchanged, and the global twin margin remains OPEN. | C | [left-divisor-signs.md](left-divisor-signs.md) |\n| `Q-lemmaV-meansquare` | PARTIAL | Does the mean-square-in-x form of Lemma V hold, and what does the proof cost? | Proved unconditionally, _H <= B(z,s)*H for every H, z and level s, and the proof stops at exactly one place, B(z,s) bounded: B(z,3.0) measures 1.3833 to 1.4883 at z = 13..37, flat to 8 per cent over the whole computable range, which is a measurement and not a bound. | none | [attack-beta2-01-lemmaV-meansquare.md](history/staging/attack-beta2-01-lemmaV-meansquare.md) |\n| `Q-lemmaV-neighbours` | ANSWERED | What do the five Lemma-V-adjacent instruments give when read at source, and which hypothesis do we violate? | The right instrument is Maynard's Lemma 6.12, owned by Deshouillers-Iwaniec 1982 Theorem 12 and sharpened by Pascadi's Corollary 18 in exactly the term that binds us; the hypothesis violated is narrower and sharper than the corpus expected, since all three papers allow arbitrary rough coefficients in three of five variables and what they require is smoothness where we have none. | none | [lemmaV-neighbours.md](history/staging/lemmaV-neighbours.md) |\n| `Q-lemmaV-usup` | ANSWERED | Does u_sup plateau below beta2, or diverge? | It rises at every added level, 2.0617 at z = 13 to 3.2026 at z = 43: a plateau below beta2 is rejected and so is the fast divergence attack 1 feared, leaving \"plateau above beta2, or divergence\", which for this route is the same verdict (divergence since REFUTED at fixed s by attack-0829n-rml-proof.md section 4.1, adjudicated in verify-0830-usup-convention.md). | none | [lemmaV-sup-extension.md](history/staging/lemmaV-sup-extension.md) |\n| `Q-level-ledger-tight` | PARTIAL | How much of the Level Ledger's 2*3^{pi(x)-1} class-discrepancy bound can be taken back? | The sup-versus-variance gap is not closed, and section 4 shows no L1 Fourier method can close it, since such a bound is already >> 2^{pi(x)} from its own j = 1 term; what replaces the exponent is an exact reduction to one finite quantity, computed by exhaustion to y = 19, and a proven CONSTANT factor 81.0 uniform in x, p and a (53.9 from the cheap y = 17 seed), which does not grow. | none | [level-ledger-tight.md](level-ledger-tight.md) |\n| `Q-lichtman-decomp` | ANSWERED | What is the wall's published constant 3.29956 made of, and is there technique slack a method of this class could spend? | The split is multiplicative, 3.2995525 = 2 x 1.649776 with the second factor an equidistribution deficit, so the brief's \"remaining 1.65 of technique\" has no referent and category (b) is empty; the floor for this class is 2 on Type-I data and still 1.122919 at unlimited level, all arithmetic SCRATCHPAD-GRADE. | none | [attack-lichtman-decomp.md](history/staging/attack-lichtman-decomp.md) |\n| `Q-lit-evans` | ANSWERED | Does Evans, Correlations of almost primes, reach y about x^0.28, does it ever fix h, and is there a secondary term to harvest for the depth law? | None of the three, and the item named the single highest-value unread item in the survey yields nothing for the depth law: over the whole 35-page extraction Buchstab, rough, sifting and omega(u) each occur zero times, so the paper has no sieve-depth coordinate, and its object is E2 numbers, a multiplicative condition rather than a roughness one. | none | [lit-evans.md](history/staging/lit-evans.md) |\n| `Q-lit-fgkt-maier` | ANSWERED | Do the corpus's FGKT and Maier quotations hold against the PDFs of record? | The quotations are VERBATIM where checked, including at the sub/superscript level HTML destroys, and the defects are in the claims built on them: a Buchstab dimension error at four sites, one of them publication-track and self-contradictory, and two provenance errors (the Maier-Pomerance equality form and the FGKT/FGKMT R-statement); Cheer-Goldston was not obtained, so one quotation stays unverified. | none | [lit-pdf-fgkt-maier.md](history/staging/lit-pdf-fgkt-maier.md) |\n| `Q-lit-fkmpt-quotes` | ANSWERED | Do the corpus's FKMPT quotations hold against the journal PDFs of record? | All three quotations are VERBATIM or CONFIRMED and two of the three surrounding claims are supported; the third is not, the cited section 1.3 does not exist (the passage is 1.1) and the connective overreaches, and five corrections are drafted including one silent elision inside a block labelled verbatim. | none | [lit-pdf-fkmpt.md](history/staging/lit-pdf-fkmpt.md) |\n| `Q-lit-halberstam-richert` | ANSWERED | Do the Halberstam-Richert and DHR citations in dhr-verification.md and two-class-lower-bounds.md check out against primary sources? | Mostly reachable, against the brief's expectation: seven claims verbatim against source including beta2 about 4.266 at the book page itself, one UNREACHABLE (HR Corollary 2.4.1), the meaning of DHR and the Friedlander attribution both settled, and 4 defects among 21 further attributions swept. | none | [lit-pdf-halberstam-richert.md](history/staging/lit-pdf-halberstam-richert.md) |\n| `Q-lit-kalmynin-konyagin` | ANSWERED | Do the corpus's Kalmynin-Konyagin citations hold against the PDF of record? | Item 2 is clean, reproducing their Theorem 1 at l_f = 2, h_f = 0, M(f) = 2 exactly; item 1 is MISROUTED, the quote being FKMPT's Definition 1 rather than K-K; and the word twin appears zero times in every version, with the calibration running opposite to expectation, since they delete two or more residue classes per prime and run a dimension-2 Mertens ledger, which our own absence rows still call ABSENT. | none | [lit-pdf-kalmynin-konyagin.md](history/staging/lit-pdf-kalmynin-konyagin.md) |\n| `Q-lit-kourbatov-grob` | ANSWERED | Do the Kourbatov and Grob quotes on which the ZONE-POSTULATE retraction rested hold against the PDFs of record? | They hold: four verification items land VERBATIM against the published article and Grob's three negatives are confirmed, so the retraction was justified and a true claim was not retracted on false evidence; the corpus under-reports Grob in two places. | none | [lit-pdf-kourbatov-grob.md](history/staging/lit-pdf-kourbatov-grob.md) |\n| `Q-lit-pdf-bft` | ANSWERED | Do our Banks-Ford-Tao quotations survive checking against the PDFs of record? | All three quotations VERBATIM, checked by scripted normalised diff rather than by eye; one was misattributed by the brief and not by the document (it is Maier-Pomerance), the demotion resting on their model R stands, and the corrections owed to the record are listed. | none | [lit-pdf-banks-ford-tao.md](history/staging/lit-pdf-banks-ford-tao.md) |\n| `Q-lit-pdf-holt-rudd` | ANSWERED | Do our Holt and Rudd citations survive checking against the arXiv PDFs of record? | Every quotation VERBATIM with no misquotation found, but Lemma 3.1 is not the lemma the demotion needs, the 2p spacing result carries no numbered theorem of that name, several attribution defects and two internal conflicts are ours to settle, and the absence claim remains an absence claim. | none | [lit-pdf-holt-rudd.md](history/staging/lit-pdf-holt-rudd.md) |\n| `Q-lit-scourfield-2008` | ANSWERED | Does Scourfield, Smooth divisors of polynomials (LMS Lect. Note Ser. 352, CUP 2008, 286-311), supply the divisor-distribution estimate that varE-theta2-proof.md section 6 leaves open? | DOES NOT DELIVER, at SECOND-HAND rung, and the chapter is still UNREAD at the page: ten channels were attempted, nine answered, and every one that answered is closed, degraded or paywalled, but the author's own restatement of the 2008 theorem, read at the page image of Scourfield, Funct. Approx. 55.1 (2016) 84 eq. (1.1), counts divisors m <= x of f(n) for n <= x with weight 1, for f a product of pairwise coprime irreducible factors each of degree at least 2, and the open step needs divisors n > 2L above the window, carrying a branch-dependent weight of total mass 4^omega, for the product of three LINEAR factors C(C-2)(C+2). Two gaps each independently fatal, plus a third that is AMBIGUOUS between two author-sourced statements and does not carry the verdict. One correction to lit-smooth-divisors.md is earned: the precision register of the owning convention is not uniformly (log y)^{o(1)} or order-of-magnitude, since (1.1) is an asymptotic with relative error O(1/log x), which is the right register and one epsilon short of the o(1/ln y) the open step needs. Step 1 of Var/E does NOT become conditional on a citation, and no absence may be written, since the chapter's interior lemmas were not read. | 9 | [lit-scourfield-2008.md](history/staging/lit-scourfield-2008.md) |\n| `Q-lit-siebert` | ANSWERED | Is Siebert's 1983 result stated for general sieve dimension kappa, or for kappa = 1 only, and does it therefore already contain the kappa = 2 extremal example that SEARCH-CONVENTIONS section 3 records as not found? | kappa = 1 ONLY, on three independent statements of the theorem (the chapter's own publisher abstract, Granville 2020 section 1, Friedlander-Iwaniec 2022 page 20), all naming the LINEAR sieve and the Jurkat-Richert f and F; the chapter's full text was NOT reached, so this is testimony not primary reading, and the SEARCH-CONVENTIONS section 3 \"no published kappa = 2 extremal example\" row needs no correction. | 0 | [lit-siebert-1983.md](history/staging/lit-siebert-1983.md) |\n| `Q-lit-wu2004` | ANSWERED | What is H_theta(2) in Wu 2004, and what does the source say about theta approaching 1? | The source does not answer it: Wu never parameterises H by a level of distribution at all, so the theta to 1 reading is not in the paper; three defects follow in what attack-lichtman-decomp.md quotes, one of them a wrong inference built on a wrong table row, and the only proven cap is H <= 1. | none | [lit-wu2004.md](history/staging/lit-wu2004.md) |\n| `Q-localized-gap` | CLOSED | Does the Localized Merge Lemma close a chain from the localized twin-slot gap to a bound on the maximal gap? | The chain does not close, and it fails on a one-line averaging argument rather than on any sieve question; what survives is the growth law of maxsum_m and the boundary question, closed at (a)/(b) = (a)/(c) = 1.0000 at every x and every m <= 1024. | none | [LOCALIZED-GAP.md](LOCALIZED-GAP.md) |\n| `Q-lp-push-x43` | ANSWERED | Does the level-D LP's DP1 > DP3 inversion survive being pushed to x = 43? | Yes, and the correction strengthens it: the two statistics still climbing at x = 23 stop climbing, the pooled floor moves UP from 3.1945 (16 readings) to 3.3152 (33 readings), DP1 reads 2.32 of the 3.2665 against DP2+DP3's 0.95, and every one of the 33 pooled readings clears the (1 + beta_2)/2 = 2.6332 the inversion needs, the tightest at 2.6692. | none | [lp-push-x43.md](history/staging/lp-push-x43.md) |\n| `Q-maier-matrix` | CLOSED | Can the Maier matrix transfer exact residue-level knowledge about twin slots into an interval statement, and if not, where does it stop? | It stops at a scale collision: at q = x# the matrix is the identity 0 = 0, at q = y# with y < x the transfer identity is PROVEN and VERIFIED 20 of 20 but is a first-moment identity only, almost-all would need the second moment which is the localized twin-gap problem itself, and Granville-Soundararajan Corollary 1.4 is vacuous at every computable scale. | none | [maier-matrix.md](maier-matrix.md) |\n| `Q-maxgap-law` | ANSWERED | Does maxgap ~ c*mbar*lnD survive far enough to contradict Maier-Pomerance, and why did two agents measure two constants? | It does not, and not because the law breaks late: it was never an extreme-value law but a curve through a two-dimensional surface, and the two agents sat on different curves without recording which; on the diagonal c really is flat over a lever 46 times long with an estimator demonstrably able to see Maier-Pomerance growth, and one repo projection is REFUTED by the audit. | none | [maxgap-law.md](maxgap-law.md) |\n| `Q-maximal-law-tpc` | ANSWERED | Is the SIEVE Gaussian maximal law TPC-implying? | YES in the only form that would deliver anything: stated for the operative remainder R_H, need/z^2 runs 0.49 to 0.61, flat and below the zone budget at every z from 13 to 43, which is the Gap Reformulation, hence the Zone Postulate, hence TPC, so no soft argument can prove it; the published above-2 theta column is not the maximal law's verdict but the law plus two lossy steps. | 0 | [phase1-T4-maximal-law.md](history/staging/phase1-T4-maximal-law.md) |\n| `Q-maxsum-growth-law` | CLOSED | What is the growth law of maxsum_m, and is there a constant C that closes the localized chain? | The chain has no chance, and for a reason cheaper than the sieve wall: the gate grows like ln x per fold while the bound it must stay under grows by at least 2.4 ln^2 x, a deficit that is unconditional in C and in the sifting parameter, and the Deficit Lemma kills the weaken-the-gate repair because the gate feeds back. | none | [localized-04-maxsum.md](localized-04-maxsum.md) |\n| `Q-maxvr-uniform` | CLOSED | Is there a level-uniform bound on max VR, and what would an all-x form need? | No level-uniform bound: all three candidate bounds are killed by the numbers, and the all-x form needs a second ingredient the Status sentence does not name, an analytic lower bound on Sbar, the count of n with both n and n+2 free of prime factors in (x, sqrt(N)]; a max VR bound is in any case not TPC-strength, and no REFUTED.md row covers the route. | none | [attack-maxvr-uniform.md](history/staging/attack-maxvr-uniform.md) |\n| `Q-measure-0904-argmax` | PARTIAL | Where inside the tile is G2(x#) attained, with what multiplicity, and how thick is the near-maximal tail? | The multiplicity half was NOT open, and is reproduced here rather than found: exact-g2-ladder.js, measure-g2z2-0829.js and ATTACKS3.md already carry it at every level to x = 43, so object-g2-read-0829.md:1000's \"unknown\" is wrong. What is new, MEASURED exhaustively over the whole period at x = 11..37 by a third independent engine: the argmax set is tau-invariant with no fixed-point hit at all eight levels and carries ZERO congruence pairs from x = 23 up (refuting this note's own F7 at four scored levels), while 58 to 82 percent of the near-maximal windows ARE exact congruence translates of another by a multiple of W/p, at 10^6 to 10^8 times the uniform null; at x = 37 the maximum is 2 positions of 7,420,738,134,810 and only 6 gaps of 217,929,355,875 sit within 10 percent of it; four of eight pre-registered forecasts failed (F3, F5, F6, F7) and two falsifiers fired (F5, F7), and the exponent is untouched. | none | [measure-0904-argmax.md](history/staging/measure-0904-argmax.md) |\n| `Q-measure-g2-generic` | ANSWERED | Where does G2(x#) sit in the distribution of the same maximum gap taken over other two-class configurations with the same census, and is the constant-shift family generic inside the free-pair family? | G2 is GENERIC in this coordinate on the registered rule, MEASURED: the midrank percentile of G2 in the census-matched two-class ensemble reads 20.00, 18.06, 26.35, 17.14 exhaustively at x = 11..19 and 13.70 sampled at x = 23, all inside the registered [5, 95] band so the falsifier does not fire, while sitting about one ensemble sd below the mean at all five levels; and the free-pair and constant-shift ensembles are PROVEN identical up to translation, so killer 1 gains no target in the class positions at these levels. | 0 | [measure-g2-generic-0829.md](history/staging/measure-g2-generic-0829.md) |\n| `Q-measure-g2z2` | ANSWERED | What is G2(x#)/Z2(x) at every level where both exist, where does the tile's least attaining position sit, and how often does a zone hold exactly two pairs? | MEASURED at 22 levels: G2/Z2 rises 1.00 to 8.1429 with its maximum 8.3214 at x = 71, is lossless only at x <= 7, and no gap of length G2 lies inside the zone for x = 11..79 (PROVEN there from Z2 < G2 plus the Zone Restriction Lemma); R0's tight case k = 2 occurs at 0 of 27,291 zones above the degenerate first one; the fourteen custody ratios reproduce digit for digit and Z2 = env at 27,292 of 27,292 zones; the pre-registration scored three of five with one pre-registered theorem refuted (the Poisson miss is 1.518e-1 over p >= 3, corrected from a mislabelled 4.177e-1); no upper bound on G2/Z2 is claimed or in evidence. | none | [measure-g2z2-0829.md](history/staging/measure-g2z2-0829.md) |\n| `Q-minus-half` | ANSWERED | Is the -1/2 anticorrelation constant of the abutting-window covariance exact, or only equidistribution-conditional? | For the flat spectral functional the -1/2 is EXACT, a two-line consequence of the mod-30 difference set, verified to relerr <= 1e-15 on the full spectrum at x = 7 and 11; the whole equidistribution question compresses exactly into two residue-class correlation sums, and the residual all-x statement reduces to G30_agg(x) < 1/2. | none | [natal-cap-26-minus-half.md](natal-cap-26-minus-half.md) |\n| `Q-mismatch-adjudication` | ANSWERED | What is the true value behind each of readings-traceability's 56 mismatches, and where does each repair land? | All 56 adjudicated with old and new recorded: 33 READING-FIXED, 5 OUTPUT-FIXED-REBOUND, 2 awaiting re-embed, 5 DOC-FIXED, 4 UNTRACED-MARKED and 7 HOLDS-AS-IS; nine were repaired by making a script produce a figure it had only described, and three of those reproductions turned up a second defect. | none | [mismatch-adjudication.md](history/staging/mismatch-adjudication.md) |\n| `Q-mobius-bv-derivation` | ANSWERED | Is there a published theorem statement of Bombieri-Vinogradov for the Mobius function at level T^(1/2)/log^B T with a maximum over coprime classes, and if not, does that estimate follow from published theorems by a derivation whose every hypothesis is checked? | No published theorem statement was located in five channels searched in the owning convention named in research/SEARCH-CONVENTIONS.md; the closest is Granville-Shao's published assertion that the result is known, which carries no proof locator for the Mobius case. The estimate is derived here from four numbered results of Koukoulopoulos, GSM 203 (Corollary 13.4, Theorem 26.2, Theorem 26.6, equation (26.3)) plus Vaughan's identity for mu, with B(A)=A+6 and an ineffective constant. This is a derivation from published theorems, not a refereed theorem, and it is read from the author's preliminary version of that book. | C | [mobius-bv-derivation.md](mobius-bv-derivation.md) |\n| `Q-mod30-tail` | ANSWERED | Are the BV survey's S1 (mod-30 tail) and S2 (fixed truncation and depth) provable as stated? | Both PROVEN at short-note grade (no second reader) once S2's freshness factor is corrected to one class per prime, which needs only P^-(m) >= q > q_i and so holds at every scour prime, not only in the prime regime; tail constant 2 ln 2 -> (ln 2)/2; nothing in the head. | 11 (retired) | [thm-mod30-tail.md](history/staging/thm-mod30-tail.md) |\n| `Q-monotonicity-sweep` | ANSWERED | Does anything in the corpus assume certificate validity is monotone in L? | One unsound artifact and one invalid inference: attack-beta2-04-loss-budget.js section 6 bisects on a predicate measured not upward-closed and is wrong at 3 of 5 levels, true first-crossings 30/72/132/174/210 against the reported 36/72/144/174/354, and redteam-DP1-certificate.js draws a global minimality conclusion from a two-point local check; corrected, worst-casing certifies within 1.00 to 2.00 of true G2 and the exponent penalty runs 0 to 0.27 and falls with x. | 3 (retired) | [monotonicity-sweep.md](history/staging/monotonicity-sweep.md) |\n| `Q-moving-cutoff-parity` | ANSWERED | Does a proof-level reading of the shifted-Mobius sieve give a valid, explicit error-tolerant arithmetic target, and does its divisor switching preserve the moving boundary? | CHECKED prior art: Murty--Vatwani already supplies the conditional squarefree/parity mechanism and fixed-residue reduction. Its printed p. 654 swap omits n+h>ey; an exact finite counterexample verifies failure of that equality. DERIVED dyadic repair keeps the boundary and gives S=C2*x-2*C2*M+D_y+O_A(x/log^A x). A one-sided D_y>=-4*x/25+o(x) on unbounded dyadic scales is sufficient, using the certified margin C2*(1-A2)>33/200. That discrepancy estimate remains OPEN. Vatwani's recovered author preprint clarifies the one-odd-exponent cases and unfilled near-level-one hypotheses. No new signed bound is obtained; literature novelty is unestablished. | C | [moving-cutoff-parity.md](moving-cutoff-parity.md) |\n| `Q-mp-derivation` | ANSWERED | Can the fold-factor field M_p be derived rather than merely measured? | Its comb structure derives: M_p is about k*W1(theta_p)*exp(-delta*theta_p/mbar_p) with W1 the exact endpoint comb at zero parameters, blind-validated at a second fresh anchor; (k, delta) remain fitted, so this is a zero-parameter derivation of the field's comb structure and not of M_p, and nothing here proves H-double-prime or touches the exponent. | none | [mp-derivation.md](history/staging/mp-derivation.md) |\n| `Q-mp-window-prereg` | OPEN | Do the M_p FORMULA bands hold at a second fresh anchor? | Pre-registration only, sealed and committed alone before the measuring producer exists: the window, the population, the zero-new-parameter formula, all 37 band pairs, three scoring criteria and the consequences are fixed here, pinned to the v1 producer's embed fingerprint; nothing is measured. | none | [mp-window-prereg.md](history/staging/mp-window-prereg.md) |\n| `Q-multiplicity3` | CLOSED | Does the m >= 3 multiplicity instrument buy anything for the anchored deficit? | Closed as an instrument: at @17 the m >= 3 statistic sees the anchored deficit worse than the pair statistic it was to replace (anchored z -0.23 against -2.71 for the overlap credit X), the Bonferroni depth ladder buys one place and dies, and multiplicity 3 in moment order buys a factor 1.3 that does not reach the anchor. | none | [attack-multiplicity3.md](history/staging/attack-multiplicity3.md) |\n| `Q-natal-cap-campaign` | ANSWERED | What is the mathematical upper limit on the number of Natal@5 slots a prime q can remove from an X-tile as the Scour drags across it? | Per-prime removal is boxed by four instruments (max overshoot +7.7% over the fair share 2N/q, worst conceivable dilation +0.7% at @17 over 92,160 tuples, certified Selberg caps 1.13-2.5x truth, and a proven prime-counting cap for q > W^(1/3)), but no per-prime cap, even an oracle, closes the first-order union bound for x >= 13 because Sum 2/q crosses 1 between @11 and @13. | none | [NATAL-CAP-CAMPAIGN.md](NATAL-CAP-CAMPAIGN.md) |\n| `Q-natal-cap-overlap-sign` | CLOSED | Is the sub-CRT overlap bound S2 <= sum of 4N/(q q') provable, per pair or in aggregate? | REFUTED as an inequality at every granularity, with 40 to 53 percent of pairs sitting above CRT at x = 11 through 23 and named exhibits at each level; what survives is proven structure, an exact per-pair-computable aggregate deficit, with the remainder a sign-alternating discrepancy at noise scale. | none | [natal-cap-12-overlap-sign.md](natal-cap-12-overlap-sign.md) |\n| `Q-natal-cap-sieve-cap` | ANSWERED | What is the rigorous sieve upper bound on how much a single scour prime can remove, and does it certify the zone? | Certified Selberg and Bonferroni caps run 1.60 to 2.36 times N over x = 7..17 with the fluctuation factor at 1.000, gross(q) hugging 2N/q to three decimals, but from x = 13 the harmonic factor alone exceeds 1, so Sum gross > N and even exact-truth caps fail. | none | [natal-cap-10-sieve-cap.md](natal-cap-10-sieve-cap.md) |\n| `Q-natal-discrepancy-lemma` | PARTIAL | Can the rotation ensemble's class discrepancy be controlled strongly enough to feed Freedman's inequality? | Lemmas 1 to 4 and Theorem 5 are stated and machine-checked (1,954 assertions, zero failures) and Proposition 6's stopped Markov-union-Freedman chain is rigorous, but none of it touches the anchored problem: the anchored path's per-step conditional variances are ensemble-typical (mean ratio 0.93 to 0.99), so the anchored side remains 100 per cent drift against the Mertens/2C2 wall. | none | [natal-cap-14-discrepancy-lemma.md](natal-cap-14-discrepancy-lemma.md) |\n| `Q-natal-onset` | ANSWERED | Does Chris's q^2 destruction-onset staircase hold, and does the onset lens open a mechanism at x = 37? | The staircase is VERIFIED over 1..10000 after aborting-on-mismatch against his three hand anchors, the three onsets collapse to two and the naive 50:50 A/B split is refuted by derivation; the x = 37 fourth lens is a clean negative and the fourth mechanism family closes. | none | [natal-onset-01.md](history/staging/natal-onset-01.md) |\n| `Q-next-correlation-source-map` | ANSWERED | Is there a primary-source statement with stronger scale or coefficient quantifiers than the ones the prime-band transfer actually consumed? | Scoped negative for the inspected statements: function-class and rate mismatches remain. The coordinating review withdraws the all-scales/dyadic-omitting requirement as a necessary gate, since bounded-interval stability yields a weak dyadic scale average from the existing input. No sufficient rate or full-coefficient theorem is matched. Proper-prime-power terms are already negligible on the fixed corner; other prime-r cofactor branches remain open. | C | [next-correlation-source-map.md](next-correlation-source-map.md) |\n| `Q-next-transfer-review` | ANSWERED | Does prime-band-transfer.md (1)-(6) follow at exactly its stated scope, and does any step fail against an adversarial reading of its hypotheses? | The derivation (1)-(6) survives at its stated continuous-average scope. The coordinating round review corrects check l: interval stability additionally gives a dyadic scale average, with a further exponent loss. There is no every-dyadic estimate, o(N), full-corner control or twin margin. Proper-prime-power branches have a separate negligible fixed-corner bound. | C | [next-transfer-review.md](next-transfer-review.md) |\n| `Q-novelty-postaudit` | ANSWERED | Of the ten results never covered by the 2026-08-13 prior-art audit, which are novel? | None survives as fully novel: two are outright prior art (one published thirteen years earlier with its headline claim contradicted by the source's own table), one is a two-line classical theorem mis-calibrated as a measurement, five are restatements or special cases of standard objects, and one names a lemma that does not exist. | none | [audit-novelty-postaudit.md](history/staging/audit-novelty-postaudit.md) |\n| `Q-np-licenses` | ANSWERED | What does NP-completeness license here, and is Brady's Problem 3 the G2 covering problem? | The corpus sentence \"his Problem 3 IS the G2 covering problem\" is false as written and must be withdrawn (one class per prime and an arbitrary finite set, against two classes at locked separation 2 on an interval); NP-completeness licenses four statements about computation and none of the five mathematical ones. | none | [attack-np-licenses.md](history/staging/attack-np-licenses.md) |\n| `Q-null-limsup` | MIXED (null-limsup-prereg.md: OPEN; null-limsup.md: ANSWERED) | Does the thinning null's maximal gap disagree with Neudecker's limsup law, which would retire the extinction half? | The trigger does not fire: the closed form is the exact law, the reduction returns Hawkins' own constant, and the real fold's record gap lands within 2 sigma of the null, so the extinction half survives and the null now supports the ceiling argument rather than threatening it; the limsup itself is not measurable and cannot be. | none | [null-limsup-prereg.md](history/staging/null-limsup-prereg.md), [null-limsup.md](history/staging/null-limsup.md) |\n| `Q-object-bridge-read` | PARTIAL | What sits between Z2(p) and G2(x#), which inequalities between them are proven and in which direction, and is the information the Gap Reformulation discards provably needed? | Map only, nothing computed: the reduction discards the origin to buy rotation invariance at a loss measured at G2/Z2 = 4.12 by p = 43, and the discarded information is PROVABLY not needed for the implication (a rotation-invariant statement, G2 < x'^2 - 2, does reach the zone statement), MEASURED to be worth about one unit of exponent against a deficit of 2.2665, and OPEN as to whether any proof needs it, with no separation theorem on record on either side. | none | [object-bridge-read-0829.md](history/staging/object-bridge-read-0829.md) |\n| `Q-object-g2-read` | PARTIAL | What is actually known about G2, at what calibration, where do the corpus's documents disagree about it, and which questions about it have not been asked? | The band is unchanged: exponent 4.26645 proven above, 2 needed, and the reading found no document that moves it; what this note adds is a variant table, a legal open set for section 4, seven drift items including one live target that is TPC-strength and unlabelled, and eight unasked understanding questions. | none | [object-g2-read-0829.md](history/staging/object-g2-read-0829.md) |\n| `Q-object-models-read` | PARTIAL | For G2(x#) and Z2(p), what does each null model in the corpus predict, which measured deviations from those models are real, and does any of them carry interval-level (wall-relevant) information? | Nothing found that carries interval-level information: seven of the nine real deviations are residue-level by the THE-LENS section 5 rule, Kourbatov's b is interval-level in form but blocked by a log-sparse quantifier and unexplained as a mechanism in both corpora, and the sub-Poisson rough-pair dispersion is classified interval-level in form and elementary in content (division with remainder, not placement) by the independent-thinning null that ran 2026-08-29 (measure-roughpair-null-0829.md), which also weakened the sub-Poisson reading since that null's own chi-squared per degree of freedom is 1 minus p rather than 1. | none | [object-models-read-0829.md](history/staging/object-models-read-0829.md) |\n| `Q-observations` | PARTIAL | Which sightings from the bench survive being stated precisely enough to check, and at what rung? | A running log, each entry carrying its own rung: the mirror-centre and divisor-landmark items stand as VERIFIED, the closed form for the half word is REFUTED as a route (max of the half equals max of the whole at every level), and the compounding window is PROVEN to close only to U = 30. | none | [OBSERVATIONS.md](OBSERVATIONS.md) |\n| `Q-obstruction-audit` | ANSWERED | Is the wall named parity the right name for what blocks G2(x#) < x'^2 - 2? | Read at source, the audit's central reason does not survive: Tao's forbidden is defined extensionally rather than intensionally, so the primality-detection step in the reduction does not put the method class outside the obstruction, and where the wall sits on our own reduction moves accordingly; the proposed text for paper/wall-note.md is left unapplied. | none | [attack-obstruction-audit.md](history/staging/attack-obstruction-audit.md), [lit-tao-parity.md](history/staging/lit-tao-parity.md) |\n| `Q-oeis-G2-submission` | CLOSED | Should the twin Jacobsthal function G2 be submitted to OEIS as a new sequence? | DO NOT SUBMIT: it duplicates A144311 (Carter, September 2008), which is G2 - 1 on the same fixed classes with no free translate and carries 22 terms to our fourteen, all fourteen agreeing; five waves missed it because every search was run on G2 and never on G2 - 1. | none | [oeis-G2-submission.md](oeis-G2-submission.md) |\n| `Q-oeis-g2-absence` | ANSWERED | Is the G2 ladder absent from OEIS at fourteen terms? | Falsified: the ladder has been in OEIS since 2008 in the run-length convention G2 - 1, as A144311 carrying 22 terms to our 14, with the search channel calibrated live in the same session on two known-positive probes. | none | [audit-oeis-14term.md](history/staging/audit-oeis-14term.md) |\n| `Q-oeis-seam-submission` | PARTIAL | Is the seam twin-pair count sequence absent from OEIS, and is the draft ready to submit? | Status DRAFT: the exact-terms absence survives on both range conventions and the 20 terms are cross-checked by two independent Miller-Rabin implementations, but qc-wave6-X section B3 found a concept neighbour, A367739, that the draft's crossrefs do not mention. | none | [oeis-seam-submission.md](oeis-seam-submission.md) |\n| `Q-ojaroudi-read` | ANSWERED | Does Ojaroudi's Zenodo preprint prove an unconditional twin-prime theorem, and is anything in it importable? | No: the fatal step imports a Selberg quadratic-form LOWER bound from Opera de Cribro Chapter 7 where none exists, and a second independent gap leaves the Kloosterman hypothesis never checked; the venue is an unrefereed self-deposit, and the only payoff is two elementary lemmas that are importable onto our tile. | none | [ojaroudi-read.md](history/staging/ojaroudi-read.md) |\n| `Q-omega-floor-blind-0830` | ANSWERED | Does the local log-slope of the pointwise floor Omega(z) of attack-0830-rec-cheapest.md (exact to z = 73, certified 2+4-chain family to z = 5e5), extended blind to new levels under a sealed forecast, behave as the DERIVED growth law Omega >> z^{16s/9}/ln^8 z predicts at s = 2.698721, and at what z would the test be decisive? | MEASURED, blind, on the derived law only: 18 of 22 sealed rows HIT; the certified 2+4-chain family at p* = (47, 43) runs 5e5 -> 1e9 (A1A2 = 8.863e17 -> 9.351e32, exact integers) with its step slope falling 4.85 -> 4.38 through every sealed band onto the law's log-corrected slope, the decisive decade 1e8 -> 1e9 reading 4.410 against 16s/9 - 8/ln z = 4.389 (HIT, kill rule not triggered at 4.452 +/- 0.011); the 4 MISSes are the ratio to z^{16s/9}/ln^8 z sitting at 2.00-2.14 above the sealed cap of 2 at z >= 1e8, the pre-declared non-adverse direction; the exact floor is extended 73 -> 113 (251 -> 1980, equal to the hill-climb floors at 89 and 101), where the law predicts nothing; no exponent moves and the law stays DERIVED and HELD. | 0 | [blind-0830-omega-floor.md](history/staging/blind-0830-omega-floor.md) |\n| `Q-operator-pair-count` | ANSWERED | Can the histogram transfer operator and the pair count be evaluated exactly at any level with no tile in memory? | Both instruments are exact and neither needs a tile resident: the operator is PROVEN and VERIFIED exactly at six folds and the pair count at every fold; the tail carries a FIT beside them that must not be used past the range it was fitted on. | none | [operator-and-pair-count.md](operator-and-pair-count.md) |\n| `Q-origin-excess` | ANSWERED | Can the Origin Excess Lemma be pushed to S = x'^2, and is the scale collision genuine or an artifact of that one argument? | It cannot, and the collision is genuine: the origin's excess is exactly a ratio of two values of the survival curve rho at the same window and two sieving levels, and the Zone Postulate asks for that ratio at u = 2, where rho takes its minimum, so no better lemma moves the boundary; three candidate mechanisms fail, each for a different reason, one of them working against us. | none | [origin-excess.md](origin-excess.md) |\n| `Q-packing-two-class-priorart` | PARTIAL | Does the two-class (twin) analogue of the Hensley-Richards packing function exist in print? | Not found, on an uncalibrated search in our own wording with no owning convention recorded, so this is the reach of the search rather than an absence; the classical one-class theory is well developed, and against A008407 the two-class ladder is bounded below by the minimal admissible 2k-tuple diameter. | none | [natal-cap-04-packing-notes.md](natal-cap-04-packing-notes.md) |\n| `Q-paired-factor-budget` | ANSWERED | Does averaging the proposed two-small-prime loss and three-small-prime gain using existing distribution theorems make their net contribution sufficient for the current twin consumer? | DERIVED from the matched classical inputs, with finite rational enclosures VERIFIED: the explicit positive family has liminf P3/(C2*x)>1, but liminf (N2-P3)/(C2*x)>11/10. Keeping only these families and the main term therefore fails at the current cutoffs. Pan--Ding supplies aggregate level x^(9/20), stronger than the earlier per-cofactor application; the complete signed margin remains OPEN. The aggregation machinery is prior art and literature novelty of this application is unestablished. | C | [paired-factor-budget.md](paired-factor-budget.md) |\n| `Q-paper-iii-restructure` | ANSWERED | Is Paper III restructured around the proven model theorem? | Yes. paper/variance-note.md now carries four new sections in place of its fit narrative: sec.8 the proven spectral form (Var/E = delta X, delta ln^2 W -> 16 C_2 e^{-2gamma}/3 = 1.109905, the Montgomery-Soundararajan main term exactly L prime by prime, so all of Var/E is discrepancy), sec.9 the decoupled model with Theorems 7 and 8 and Lemmas 9-11 written out with proofs and the closed form lambda_2(2) = 1 - e^{-2gamma}(9/2 - 4 ln 2) = 0.45546 plus the theta = 1 identity with Gorodetsky, sec.10 Conjecture 1 (delta(X - X_dec) -> 0) with eight exact ratios, two of three lag groups closed and the third named as a divisor-distribution statement not yet searched, and sec.11 the 0.611 refutation as a short calibration paragraph; sections 1-5 are untouched, the abstract and section 6's open question are rewritten to match, and the \"u = 2 exactly\" defect in section 7 is corrected. | 9 | [paper-iii-restructure.md](history/staging/paper-iii-restructure.md) |\n| `Q-paper-kk-draft` | ANSWERED | Is the K–K two-class lower bound written up as a refereeable draft? | Yes. paper/kk-lower-bound.md, self-contained at INFERRED grade, carrying Theorem B (the substitution), Theorem A (the published-ingredient x ln x chain) and Proposition 2 (the free G2 >= g transfer), with every constant, every consumed hypothesis, the y0 = 10^134.1 floor, the three residuals, and a per-number provenance appendix; nothing in it is new mathematics and nothing in it is refereed. | none | [paper-kk-draft.md](history/staging/paper-kk-draft.md) |\n| `Q-parity-adversary` | ANSWERED | Does a parity adversary built on our own stretches survive at every level, and what precision class does a certificate need? | It does NOT survive: the exact-data adversary dies at D* about Q^1.18 (MEASURED, exact arithmetic, 43 anchors Q <= 200), and the certificate that kills it is itself destroyed by 0.30 of one count per modulus, so Z2's precision class is now measured rather than described; it is PROVEN not to be a proof ingredient. | Z2 | [attack-parity-adversary.md](history/staging/attack-parity-adversary.md) |\n| `Q-parity-dstar-law-0829n` | CLOSED | Does the exact-data parity adversary's killing level D*(Q) follow D* ~ Q^c with a stable c on the decades past Q = 200, and is c distinguishable from the independent-thinning null's? | No law in Q and CLOSED by the sealed kill rule (score 5 HIT, 1 MISS): MEASURED on 271 anchors 211 <= Q <= 13679, the Q-slope moves 1.446 to 1.204 to 0.575 across ranges while ln D* on ln width holds slope 1.097 +- 0.012 with a quarter of the scatter, D*/width about 0.5, and the independent-thinning null reproduces the width law within 0.025 in slope; the mechanism is that D* is where a modulus first reads a single position of the window, which is the wall survey's remainder statement in the stretch coordinate and names nothing new. | Z2 | [attack-0829n-parity-dstar.md](history/staging/attack-0829n-parity-dstar.md) |\n| `Q-perfold-error-model` | ANSWERED | What is the per-fold error model behind the extinction law's +/-3 sqrt(lambda) failure at 43.2 per cent? | X_p ~ Poisson(lambda_model(p, Y) M_p) with M_p a deterministic factor of the fold, the same number at every window length from 2e7 to 2e11 and at every anchor, MEASURED and not derived, and it holds on a sealed blind test at a fresh anchor. | none | [perfold-error-model.md](history/staging/perfold-error-model.md) |\n| `Q-perfold-window` | OPEN | Does the negative-binomial per-fold error model's band hold at a fresh blind anchor, after the extinction law's +/-3 sqrt(lambda) clause failed at the fifth window? | Pre-registration only, committed alone before the producer existed: the population is fixed in advance at the 37 folds p >= 100 with predictive mean or lambda_model at least 1, and the bands are printed verbatim from the model's embedded Stage 4; nothing here is measured. | none | [perfold-window-prereg.md](history/staging/perfold-window-prereg.md) |\n| `Q-phase1-W1b` | ANSWERED | Which unapplied wave-6 findings on the research notes were still live, and what did applying them change? | Most of wave 6 was already applied, so the yield is the residue plus one code defect nobody had fixed: a3-05-bound-L.js's truncated cyclic replay was printing a FALSE PASS on kappa(m) <= L+2 at fold 7, and fixing it kills a claim; the gate reads 0 findings across all seven checks at hand-back. | none | [phase1-W1b-applied.md](history/staging/phase1-W1b-applied.md) |\n| `Q-polylog-fold-transfer` | ANSWERED | Can the fixed-fold calculation be made uniform at growing depth, and transferred to the actual bilinear remainder with an adequate error? | The local bilinear model is O_H(x/log^H x) uniformly for polylogarithmic squarefree moduli. Its absolute reconstruction error is at least (1/2+o(1))*x. The needed signed reconstruction bound remains OPEN, so this model does not establish the twin-prime lower bound. | C | [polylog-fold-transfer.md](polylog-fold-transfer.md) |\n| `Q-prime-band-completion` | ANSWERED | Can the remaining squarefree correlation be localized to simpler prime and harmonic bands, and does completion then permit a direct general bilinear Kloosterman estimate? | The pilot reduces, with power-saving endpoint errors, to p in (x^0.235,x^0.24], q in (x^0.045,x^0.05] and positive h in (x^0.029,T], with T=O(x^0.0342). Both primes exceed 2h eventually. Completion is exact but produces coupled h-dependent coefficients on every dual residue modulo q. Its full-frequency Kloosterman matrix has norm exactly q, so a uniform power saving by an arbitrary-coefficient operator bound is impossible at that scope. This completion alone leaves the 61/20000 budget deficit; prime-dispersion.md controls the pilot using a different second moment. No twin lower bound is established. | C | [prime-band-completion.md](prime-band-completion.md) |\n| `Q-prime-band-transfer` | PARTIAL | Can the nonmultiplicative prime-band weight be represented by bounded multiplicative functions, and what does an available quantitative correlation theorem actually give after all scale and normalization costs? | A smooth Fourier superposition represents mu(n)L_B(n)/log X exactly on n<=2X+2. The quantitative non-pretentious input and a mesh give a continuous scale-average saving for the exact s=s'=1 windows. The round review additionally derives a dyadic scale-average saving by bounded-interval stability, with exponent c_*/2. Neither estimate gives an every-dyadic bound, o(N), full-corner control or a twin margin. The extension in cofactor-progression-transfer.md now estimates a growing polylogarithmic cofactor family; the full remainder remains open. Proper-prime-power branches have the separate bound in corner-correlation §1.1. | C | [prime-band-transfer.md](prime-band-transfer.md) |\n| `Q-prime-detection-inputs` | ANSWERED | Can we give a locally correct positive comparison sequence, justify the ordinary arithmetic inputs, and prove exactly what one additional estimate would buy? | Yes: the comparison sequence and Type I estimates are derived, and a classical Vaughan decomposition gives S(x)=C2*x+B(x)+O_H(x/log^H x) for every fixed H. A fixed one-sided saving over B>=-C2*x+o(x) implies twins on unbounded scales. This is a fully specified classical reduction, not a new cancellation estimate or proof mechanism. | C | [prime-detection-spec.md](prime-detection-spec.md) |\n| `Q-prime-dispersion` | ANSWERED | Can a second moment retaining two right primes control the remaining localized squarefree pilot after its actual conductor and coefficient costs are included? | Yes at this rectangle: aggregation in m=dp before Cauchy gives a logarithmically bounded coefficient on a long interval. The only zero-phase pairs are identical (q,h). Distinct primes give modulus e*q1*q2 with gcd loss at most O(A*Z); completion and Weil bound all off-diagonal terms. The diagonal has exponent 1989/2000 and the largest off-diagonal exponent is 1173/1250 before an arbitrarily small loss. The remaining high-frequency polynomial is O(x^(199/200)); consuming the previous reductions gives R_IJ=O_H(x/log^H x) at D=x^0.277, E=x^0.467. This is a full rectangle at product scale x^0.744, not a uniform product cutoff or a twin lower bound. | C | [prime-dispersion.md](prime-dispersion.md) |\n| `Q-prime-power-dispersion` | ANSWERED | Does the repeated-prime structure control the BB contribution beyond its two-norm bound, and can a prime-power moment include every harmonic uniformly? | The decomposition leaves an exceptional coefficient with norm at most sqrt(D) times logarithms; its odd prime-power core has a derived all-harmonic dispersion bound. The same bound for first powers removes the former left-prime cutoff and unpaired diagonal constraint. Two strict inequalities control a region including d~x^0.247, e~x^0.612 at product scale x^0.859. Their product-exponent supremum is 5757/6700, not a uniform product cutoff or a twin lower bound. Exact finite checks run separately from this classical-input derivation. | C | [prime-power-dispersion.md](prime-power-dispersion.md) |\n| `Q-prior-art-audit` | PARTIAL | What in this programme is rediscovery, and what might be new? | Rediscovery, prolific and correct, spanning 170 years (Smith 1857 to Tafula 2015); three candidate novelties survived the 2026-08-13 audit and item 1 was cut back on 2026-08-18 by OEIS A144311 (Carter 2008), and several negatives were run in our own wording rather than the owning convention, which is worth little. The 2026-09-06 global smooth majorant uses an explicitly matched prior-art finite-difference mechanism; literature novelty of the complete application is unestablished. | none | [PRIOR-ART.md](PRIOR-ART.md) |\n| `Q-prior-art-last-ground` | ANSWERED | Is MathSciNet reachable, and does the two-class upper-bound absence survive a complete title enumeration? | MR Lookup answers author/title/journal/year queries free (bibliographic fields only, no review text or MSC), so SEARCH-CONVENTIONS section 5 needs correcting; the standing sentence survives, since the complete MathSciNet title record for Jacobsthal contains every published upper bound and every one is one class per prime, and Holt's 2022 book is in no bibliographic database. | none | [attack-prior-art-last-ground.md](history/staging/attack-prior-art-last-ground.md) |\n| `Q-proposals-prior-art` | ANSWERED | What is the prior-art risk for the six candidates in the paper-proposals registry? | None is clean: three are ADJACENT with the boundary drawn Holt-style (one of them breaking a sentence the corpus rests on), the thinning null is PRIOR ART FOUND under Hawkins' random sieve with the Fold Moment Identity NOT FOUND, and only the reverse-direction Suen import reads NOVEL-SO-FAR; every verdict names what was not searched. | none | [proposals-prior-art.md](history/staging/proposals-prior-art.md) |\n| `Q-qc-arch` | ANSWERED | Is the corpus's architecture, ordering and file layout readable by a fresh agent at minimum context cost? | Diagnosis only, no existing file edited: the restructure plan of section 2 was accepted over the shepherd's, and the leading verdict is to leave OBSERVATIONS.md exactly as it is, in place, unsplit and off the reading path, taking out only four inbound citations that name it as the home of something it does not own. | none | [qc-arch.md](history/staging/qc-arch.md) |\n| `Q-qc-campaign` | ANSWERED | What did the 2026-08-17 consistency campaign run, find, decide and apply? | The shepherd's ledger: Chris's duplication rule and splitting ruling recorded in his words, the qc framework and the provenance index delivered in wave 1, and every partition's results with the decisions that settled them, including two novelty verdicts flagged as publication-track changes. | none | [qc-CAMPAIGN.md](history/staging/qc-CAMPAIGN.md) |\n| `Q-qc-checks-9-10` | ANSWERED | What do the provenance and search-convention gate checks find, and why are they held out of the default gate? | Both are built, calibrated and in the selftest but deliberately behind --pending because they fire on live claims an attack agent is mid-flight against: provenance reads 8 findings and search-convention 75 on the 2026-08-18 corpus, the flip to default is one line, and what could not be mechanised is stated. | none | [qc-checks-9-10.md](history/staging/qc-checks-9-10.md) |\n| `Q-qc-compound` | ANSWERED | Which named claims are compounds whose parts carry different calibrations, and how should each be decomposed? | Twelve compounds decomposed and eight examined and left whole, with about 90 corpus sites needing rewording; the test each decomposition must pass is that Proven: becomes unwritable rather than merely wrong, and the sharpest finding is that natal-cap-28's fully analytic certificate law has no markdown home at all, which is why its status is unfindable. | none | [qc-compound.md](history/staging/qc-compound.md) |\n| `Q-qc-history-migration` | ANSWERED | Which body text is history and should move to CHANGELOG.md or history/? | The headline is a negative and it is the most useful thing in the report: no body file is a session record that should move wholesale, so what moves is per-section, and the pass leaves an ORPHANED-CLAIM list of live errors plus per-file findings over all 63 body files. | none | [qc-history.md](history/staging/qc-history.md) |\n| `Q-qc-numbers` | ANSWERED | Do two places in the corpus ever give the same number for the same object, and does each remote quotation carry the caveat its home document requires? | 16 findings organised by quantity, led by the rho disagreement between U-FRAME:646 and gate-multiplies:360; diagnosis only, no existing file touched, and arithmetic-per-number is audit-numbers.md's axis rather than this pass's. | none | [qc-numbers.md](history/staging/qc-numbers.md) |\n| `Q-qc-papers2` | ANSWERED | Do the paper suite's five doors and four faces restate their research homes correctly? | Fifteen findings in sections 7 and 7A plus a suite sweep: a door resting entirely on a column its own cited file disclaims as invalid, whose corrected computation reverses the reading; a degree-4 certificate credited with a result needing moments through order 6; and the same paper contradicting itself forty lines apart on the anchored calm. | none | [qc-papers2.md](history/staging/qc-papers2.md) |\n| `Q-qc-queue-9-10` | ANSWERED | Can the provenance and search-convention queues be cleared without suppressing anything? | TOTAL 84 to 0 with nothing suppressed and every finding resolved in the corpus: all six not in OEIS claims were FALSE, since the G2 ladder is A144311 + 1 and PRIOR-ART.md already recorded the find 270 lines further down; the honest remainder is that five of 75 were resolved by rewording a sentence that was never a literature claim. | none | [qc-queue-9-10-worked.md](history/staging/qc-queue-9-10-worked.md) |\n| `Q-qc-refs` | ANSWERED | Are the corpus's external citations and its quotations sound? | Nine dead quotations and eight wrong external citations, the worst being that the corpus sits on the retracted FKMPT constant, an OEIS crossref off by one twice in a document addressed to an OEIS editor, and a false premise (\"Kanold, Stevens and Paseman all give exponent 2 + eps\") under an open TODO item. | none | [qc-refs.md](history/staging/qc-refs.md) |\n| `Q-qc-scope-R` | ANSWERED | Where does the summary layer restate a result with the scope its home document attaches dropped? | Fifteen findings, 2 HIGH, 3 MEDIUM-HIGH, 4 MEDIUM, 6 LOW: one HIGH is a false statement in README.md Status introduced by this campaign's own wave 2, the other is both papers giving a false reason for the X-limitation theorem's scope, and the coverage statement at the end is short and matters. | none | [qc-scope-R.md](history/staging/qc-scope-R.md) |\n| `Q-qc-scope-T` | ANSWERED | Does the surface partition R left unswept carry the campaign's dominant defect, a scope dropped in restatement, or a false absence claim? | Fifteen ranked findings plus the TA, TB and TC blocks: a scoreboard still logging the rich vein as the campaign's one unexplained novelty after a same-day follow-up solved it, a fourth copy of the retired sifting limit, and a largest-single-event superlative refuted by the event log it is read from; adjudication only, nothing edited. | none | [qc-scope-T.md](history/staging/qc-scope-T.md) |\n| `Q-qc-scripts-S1` | ANSWERED | Do the 37 natal-cap scripts' READINGS blocks hold against their own output and against the documents citing them? | Twenty-six of 37 are consistent and eleven carry a defect, none of them contradicting the summary layer on a status claim: a false promise of a record with no READINGS block, a reading refuted by its own printed output, an absence claim at the site a whole wave read as absolute, and a Miller-Rabin bound wrong by nine orders (3.186e14 for 3.186e23) at three sites. | none | [qc-scripts-S1.md](history/staging/qc-scripts-S1.md) |\n| `Q-qc-scripts-S2` | ANSWERED | Do the fold-profile and attack script families say what their own outputs say? | Thirteen defects over 34 scripts and 5,673 lines, five of them HIGH: an untested identification two later scripts had already killed, a printed reading refuted by its own table, a section that has printed only its header since it was written, and a reproduction pointer to a scratchpad that is gone; script-side ones fixed, document-side ones left as evidence, gate 0 on all seven checks. | none | [qc-scripts-S2.md](history/staging/qc-scripts-S2.md) |\n| `Q-qc-scripts-S3` | ANSWERED | Do the 57 non-campaign research scripts' banners, embedded OUTPUT blocks and READINGS agree with each other and with the documents citing them? | Defects ranked by consequence, the leading one a3-02-diagonal-f.js carrying the refuted L scanner while naming as its source the sibling that had been repaired underneath it, the two state machines disagreeing at exactly one cell, L(T23, 29) = 3 against the true 2, now fixed; the partition ends with qc.js at 0 on all seven checks and audit-numbers.js 78/78. | none | [qc-scripts-S3.md](history/staging/qc-scripts-S3.md) |\n| `Q-qc-shepherd` | ANSWERED | What does the wave-1 shepherd find on the corpus's own structure? | The largest structural defect is that U-FRAME is organised by campaign rather than by claim, with five whole sections of process record sitting in working documents, a PROVEN lemma quoted without the hypothesis that makes it non-vacuous, and three documents disagreeing about how many hypotheses that lemma has; the claim-status register across named objects was not covered and is named as the highest-value remaining axis. | none | [qc-shepherd.md](history/staging/qc-shepherd.md) |\n| `Q-qc-slopes-K` | ANSWERED | Are the three fits at slopes 1.062, 1.451 and 1.2992 numerically inconsistent? | There is no numerical inconsistency: the seven tile points are seven of the 42 census points with f agreeing to printed precision, restricting the 42 to the seven's x reproduces 1.062 exactly, and the two estimates are statistically indistinguishable (seven-point 95 per cent interval [0.32, 1.81]); but the corpus's stated reason is wrong, the comb does not explain it, and the coefficient is range-dependent under every specification, so no single slope exists. | none | [qc-slopes-K.md](history/staging/qc-slopes-K.md) |\n| `Q-qc-status` | ANSWERED | Is any named object's claimed status inflated, contradicted or misattributed? | Thirteen findings, 12 with replacement text: the worst scope inflation is the Fused-Window Calm Lemma listed as proven when two of its four legs are not, and \"the matching lower bound\" claims a sharpness the corpus explicitly disclaims; on attribution the corpus is in better shape than the brief assumed. | none | [qc-status.md](history/staging/qc-status.md) |\n| `Q-qc-transfers-J` | ANSWERED | Are the thirteen transfers findings real defects or benign restatements? | Adjudicated by word diff with the diff pasted under each finding: 11 KEEP, 2 DEFECT, both in gate-multiplies.md and both one root cause, 0 false positives, and nothing needing compute or a human decision. | none | [qc-transfers-J.md](history/staging/qc-transfers-J.md) |\n| `Q-qc-wave6-U` | ANSWERED | Do ATTACKS.md and ATTACKS2.md say what the corpus above them says? | Fifteen findings on 57 lines, led by a campaign verdict awarding theorem-grade to a curve its own home marks MEASURED and Hardy-Littlewood-conditional, the rich vein still logged as an unexplained novelty, and the anchored-window programme called unexplored where a research note and a 34 kB paper exist; six of the fifteen turn on a frozen-record-versus-current-state decision nobody has made. | none | [qc-wave6-U.md](history/staging/qc-wave6-U.md) |\n| `Q-qc-wave6-V` | ANSWERED | Do the sixteen fold-profile scripts carry auditable evidence, and what do they show when run? | All sixteen were run and now carry a real OUTPUT block and a numbered READINGS block, each verified by re-running and diffing; eighteen findings follow, the highest being a lemma quoted as proven that is false as written in two documents while three scripts in this partition print the counter-example. | none | [qc-wave6-V.md](history/staging/qc-wave6-V.md) |\n| `Q-qc-wave6-W` | ANSWERED | Do the three expensive natal-cap artifacts reproduce? | All three re-run, after the complete @41 march log turned up on disk outside the repo in a place no wave had cited; a three-way tally per file, with the .md disagreements logged rather than edited and the coverage limits stated. | none | [qc-wave6-W.md](history/staging/qc-wave6-W.md) |\n| `Q-qc-wave6-Y` | ANSWERED | Do the eight run logs, the three committed .txt outputs and the leaf-note residue say what the corpus above them says? | Findings on all three surfaces, including a stated bound on sup\\|R\\|/(H*M) that the log's own full-period block violates by 5.6x, an S6 block produced by a function the repo has since flagged wrong while the log stays unflagged, and a false absence claim in G2-STATE that the same document refutes 162 lines later; line numbers were taken against a tree being modified concurrently. | none | [qc-wave6-Y.md](history/staging/qc-wave6-Y.md) |\n| `Q-qc-wave6-absence` | ANSWERED | Do the corpus's [ABSENT], [INFERRED] and [UNVERIFIED] markers survive a check against fetched primary sources? | First web-enabled pass, no repository file edited: the exact-terms OEIS absence for the seam count sequence survives on both range conventions but a concept neighbour, A367739, goes unmentioned in the draft's crossrefs, and the note closes with an explicit list of what it could not reach, led by the MathSciNet citation graph of Iwaniec 1978. | none | [qc-wave6-X.md](history/staging/qc-wave6-X.md) |\n| `Q-quadpoint-blind-31607-0830` | ANSWERED | Does the zero-parameter depth law y*(Q), stated on the set where T(Q) >= 1 and on both the 1/(2e^gamma) = 0.280730 and the u*omega(u) = 2 (1/u* = 0.280438) forms with and without the finite-size Mertens factor, survive a sealed blind test on the bands past Q = 10007, and can that decade tell the two asymptotes apart? | MEASURED, blind, and it decides nothing asymptotic: on 8363 new anchors 10009 <= Q <= 100003 the 76 sealed rows score 64 HIT and 12 MISS with every MISS on the two naked asymptotes as pre-declared, the two finite-size zero-parameter comparators HIT every row at every band (residual on ln y*/ln h -0.00022 to -0.00002 against C2), K*/pool falls 0.032 -> 0.026 -> 0.020 -> 0.016 -> 0.013 with no bend and a quiet sup, and the band means sit 0.0017 to 0.0037 below both asymptotes, six to twelve times their 0.000291 difference, so which limit the deficit tends to is set by the comparator and not by the data. | Z2 | [blind-0830-quadpoint-31607.md](history/staging/blind-0830-quadpoint-31607.md) |\n| `Q-quadpoint-c1c2-0830` | ANSWERED | Does the fourth decade's post-hoc, unsealed preference for the omega-form comparator C2 over the exact-Mertens comparator C1 (C1 above the data by 3.46 and 8.83 se at B11 and B12, C2 within 2.82 se) survive a sealed blind test at the fifth decade, Q in (100003, 316243], and what does that say about which mechanism explains the depth law's drift? | MEASURED, sealed before the run and scored blind on 17702 new anchors, and it decides nothing asymptotic: the sealed D1 row HITs at both B13 and B14 (C1 above the data by 13.33 and 23.82 se, C2 within 1.26 and 1.82 se), so by the sealed rule the omega-at-finite-u crossing of import-rough-anatomy.md §2.1 is CONFIRMED over the exact-Mertens product of u2-engine-depth.md §5 as the depth law's main term at the sampling scale, while both still HIT every ±0.0015 row, the kill rows on C2 all HIT, K*/pool falls 0.013 -> 0.010 -> 0.008 with no bend, and the band means sit 0.0012 to 0.0018 below both asymptotes, so the two constants 0.280438 and 0.280730 stay unseparated by the pre-declared rule and the run separates two mechanisms, not two limits. | Z2 | [blind-0830-quadpoint-c1c2.md](history/staging/blind-0830-quadpoint-c1c2.md) |\n| `Q-quadpoint-prior-art` | ANSWERED | In the owning conventions, who holds the rough-pair census, the capture identity and the crossing law, and is any of it ours? | The object and the identity are standard (the dimension-2 twin sifting function plus Buchstab's identity; ADJACENT-STANDARD, no verbatim carrier, no novelty language); only the comparison X(y) < T and the crossing law stay ABSENT-PER-CONVENTION, and the Buchstab correction is 6.6% of the top-band residual, not 2%. | Z2 | [quadpoint-prior-art.md](history/staging/quadpoint-prior-art.md) |\n| `Q-quadpoint-transplant` | ANSWERED | Does the anchored-cap family transplanted onto the stretch window S_Q certify every anchor, and what does the depth cost do with height? | Pre-registration for the decade extension, committed ALONE before the segmented producer existed: it fixes the FALL trend line, the registered band readings, the kill rule and a digit-exact calibration gate that must abort before any new figure prints, and states that a falling K*/pool over two decades would still be only a measurement. | Z2, Z1 (retired) | [attack-quadpoint-01.md](history/staging/attack-quadpoint-01.md), [attack-quadpoint-02.md](history/staging/attack-quadpoint-02.md), [quadpoint-decade-prereg.md](history/staging/quadpoint-decade-prereg.md) |\n| `Q-quartic-fourth-moment` | PARTIAL | Can the fourth moment of rho-tilde, attack-rhoms-01's named stopping object, be bounded? | Not yet: the genuine quartic correlation is small (OFF/m2^2 measured -0.0454, -0.2767, +0.0101, -0.0446 at z = 13..23), the resonance condition is proved to be a local covering condition, and the whole question transfers by two exact identities onto a Lambda(4) inequality for one explicit cosecant sum whose constant is flat; the new increment rung clears z = 47. | none | [attack-quartic-01.md](history/staging/attack-quartic-01.md) |\n| `Q-reachability-coverage` | ANSWERED | Which part of the original divisor domain can no improvement to the grouped moment's nonzero kernel reach, in either orientation, and what would a uniform saving on that kernel actually buy? | For a prescribed fixed exponent margin eta0, the zero budgets leave S_0={p<2eta0,q<2eta0} below that required margin regardless of the nonzero-kernel saving. Uniform gamma=2 covers its complement at that prescribed margin. Smaller positive margins can reach fixed interior points of S_0; only p=q=0 remains at zero budget one for every margin. The proposed eta0~loglog x/log x substitution is not justified by estimates carrying x^epsilon losses; its box count is geometry only. The sufficient twin margin remains OPEN. Independent review F3 and F5 correct the quantifiers and the prime-power tail proof; round-review-0906 corrects the diagonal scope, fixed-margin bulk claim and six-cut separation. | C | [reachability-coverage.md](reachability-coverage.md) |\n| `Q-readings-traceability` | ANSWERED | What is the source of each of the 1,592 figures that appear in a script's READINGS and not in its own OUTPUT block? | The running record: nothing in the pass changes a number, 56 mismatches are classified in eight kinds for adjudication elsewhere, and the advisory count rises when the work is done right because a provenance declaration is prose that carries printed values. | none | [readings-traceability.md](history/staging/readings-traceability.md) |\n| `Q-rec-cheapest-0830` | PARTIAL | At the cheapest legal point of REC(s, u0), s -> 1+sqrt(e) and u0 -> beta_2, where the saving asked over the triangle inequality is only z^{1.031}, does any of the five dead routes close, does a mixed mean-square-plus-cap strategy close, and is the failure a proof gap or a truth gap? | No route closes and the failing step is yesterday's statement, so PARTIAL on the routes; but the step is located as a TRUTH gap, not a proof gap, asymptotically: the certificate's own value at a CRT-planted doubly-smooth window is cc(r) = -(A1A2 + A1B2 + B1A2) with every term a count of exit chains of the Rosser supports (PROVEN, checked exhaustively at z <= 73), the window bound T <= cc(r) + H - 1 is exact, and the exit-chain count is DERIVED (held) to grow like z^{16s/9 - o(1)} at four primes and z^{2s - o(1)} in the limit, so REC(s, u0) is false for every u0 < 16s/9 = 4.7088 at s -> 1+sqrt(e), the whole legal band (2, beta_2] included, and RML(alpha) with it; at every computable z it is a proof gap (Omega(z) exact to z = 73 reads 251 against z^{u0} = 1.05e6, and a certified lower bound 8.86e17 at z = 5e5 sits a factor 1.2e6 under z^{u0}); no exponent moved. | 0 | [attack-0830-rec-cheapest.md](history/staging/attack-0830-rec-cheapest.md) |\n| `Q-recon-0828-covering` | ANSWERED | Does the interval-covering, automated-proof or extremal-combinatorics literature hold a method that bounds the length of an interval covered by two residue classes per prime, or that turns the exact G₂ ladder into a statement at all x? | No route. Fifteen angles examined, fourteen dead on arrival or already closed here; the best interval-covering bound in print is Crittenden-Vanden Eynden's own Lemma 2, which reads 1.02e12 against G₂(79#) = 1710 and is 8181x weaker than the corpus's own 4.2665 sieve bound at the same level; the one survivor, transplanting Costello-Watts's recurrent certificate to two classes, is a computation that certifies finitely many levels and whose analytic half is Erdős #970 at one class. | 0 | [recon-0828-covering.md](history/staging/recon-0828-covering.md) |\n| `Q-recon-0828-farfields` | ANSWERED | Do any of the fields still absent from IMPORT-MAP.md hold a theorem whose hypothesis the tile satisfies exactly and whose conclusion is a lower bound on a periodic set's count in every window, a maximal-gap bound for a product-structured periodic set, or a positivity statement at a distinguished point? | No route from any of the sixteen candidates priced; two reach STRONG-ANALOGY or better and CLEAN and both bank only a closure or a wall address, the Fourier one by a one-line non-vanishing proof that empties the Donoho-Stark/Tao/Meshulam family on this object, and Janson's lower tail by saturating at exp(-1/c) >= 0.0805 against the 1/W a position union bound needs, short by 3.6e2 at x = 11 and 2.8e13 at x = 41. | 0 | [recon-0828-farfields.md](history/staging/recon-0828-farfields.md) |\n| `Q-recon-0828-jacobsthal` | ANSWERED | Does the 1978-2026 literature on upper bounds for Jacobsthal-type functions and maximal gaps in sieved sets hold anything that lowers G2 or Z2? | No route. Thirteen angles graded, ten dead on arrival (eight already closed in this corpus, two on their own structure), three surviving and none of them lowers an exponent: a kappa=2 extremal candidate built from an exceptional character rather than from Liouville, which can only close the band faster; an unread 1983 Siebert chapter that could already contain it; and one published anchor. Z2 has no non-TPC band at all. | none | [recon-0828-jacobsthal.md](history/staging/recon-0828-jacobsthal.md) |\n| `Q-recon-0828-rough` | ANSWERED | Does the 2000-2026 literature on y-rough numbers, sifted sets and short intervals hold any theorem that lowers Z2 or G2, meaning any control on the largest gap or the minimum count of a sifted set in EVERY short interval, or any structure of rough numbers a gap bound could use? | No. Ten angles examined, ten killed for this purpose; the strongest every-interval statement in print for one-class rough numbers is Iwaniec's S(x,y,z) >= (4y/log^2 y)(log(y/z^2) - O(1)) for y >> z^2, uniform in x, giving J(P(z)) << z^2, and it is the linear sieve run exactly at its sifting limit beta_1 = 2 with no slack outside it; no two-class analogue exists in print at any exponent, the corpus's own x^4.2665 is beta_2 by the same mechanism, and every remaining branch of the field (Matomaki-Radziwill, Gorodetsky, Friedlander-Iwaniec Cor 6.28, Gafni-Tao) carries the almost-all quantifier, which a second-moment statement can never convert into a bound on one empty window. | Z2, 0 | [recon-0828-rough.md](history/staging/recon-0828-rough.md) |\n| `Q-recon-0828-sieve` | ANSWERED | Is there anything in the 2008-2026 sieve literature, or in the parity-breaking literature, that lowers beta_2 = 4.26645 or supplies a consumer for a level theta > 1 on the two-class comb? | No route. Twenty-one angles examined, nineteen dead on arrival, nothing in print beats 4.26645 at kappa = 2 at any level on four calibrated channels; the two survivors are an owning-convention row and one unresolved hypothesis in a five-page 1990 announcement whose stated requirement beta < 3 excludes beta_2 = 4.26645. | 0, Z2 | [recon-0828-sieve.md](history/staging/recon-0828-sieve.md) |\n| `Q-recon-0829-escapes` | ANSWERED | Each of the three killers is an obstruction shape that occurs elsewhere in mathematics; where has each shape been beaten in print, by what mechanism, under what hypothesis, and does that hypothesis hold for the tile and zone objects? | Of nineteen escapes read, fourteen fail a hypothesis check (twelve on a fact already in this corpus), four fail some other way and one (the decoupled vector sieve) stays OPEN as the programme's existing road, so no route opens; Granville-Soundararajan's uncertainty principle is NOT killer 3 in print, is adverse-direction, and is already cited here twice. | 0 | [recon-0829-escapes.md](history/staging/recon-0829-escapes.md) |\n| `Q-recon-0829-farfields2` | ANSWERED | Do any of eight fields still absent from IMPORT-MAP.md and from the three 2026-08-28 recon notes hold a theorem whose hypothesis the tile or the zone satisfies exactly and whose conclusion is a lower bound on a periodic set's count in every window, a maximal-gap bound for a product-structured periodic set, or a positivity statement at a distinguished point? | No route from any of the eight; the pass banks one PUBLISHED-ANCHOR (Holt and Rudd's transfer matrix has its full spectrum in print, second eigenvalue a2 = prod_{17<=q<=p}(q-3)/(q-2), printed 0.10206751799779 at p = 999999999989, decaying only like 2.82/ln p, and this corpus's prior-art record names the eigenstructure and the binomial eigenvectors but carries the per-prime eigenvalue (p-j-1)/(p-2) at lit-pdf-holt-rudd.md line 259 but no product-over-primes closed form, no numerical value and no rate), one CLOSURE with mechanism (Beurling-Selberg is the proof device behind the large sieve, whose constant N-1+1/delta is an l2-to-l2 statement and whose error 1/delta exceeds the window at the operative theta_total > 2, so the whole extremal-function family reproduces a row already rejected), and three wall addresses; two angles are TPC-strength at the form that would pay and are labelled (ii). | 0 | [recon-0829-farfields2.md](history/staging/recon-0829-farfields2.md) |\n| `Q-recon-0830-rec-killrun` | ANSWERED | Does the literature hold, in its own conventions, a theorem whose hypotheses REC(s, u0) (the sup-versus-rms recovery of the level-D signed remainder over all positions, attack-0829n-rml-proof.md section 3) or the weighted kill-run K*(s) (the longest run of level-s slots the primes in (s, 2s] can kill, attack-0829n-doubling-bridge.md sections 0 and 3) satisfies as stated, or a Maier-type theorem that makes REC false for two-class sifted sets at some u0 below beta_2? | No theorem applies to either object as stated, on four calibrated channels searched in the owning conventions; every neighbour is graded NEAREST with its unmet hypothesis named, and neither object is on any refuted row. On the Maier question the answer is negative for REC as stated (the Maier family bounds the COUNT, i.e. the full-level remainder, never a level-D truncation) but carries one calibration: at one class and full level the REC-shaped inequality is FALSE asymptotically, by Buchstab's origin ratio against the Montgomery-Vaughan full-period variance ceiling, and the embedded arithmetic puts the level where that falsity first shows at log10 z between 16 and 141 depending on u0 and epsilon, so a finite-z margin of the kind the corpus measures cannot see a failure of this type; at two classes the same full-level statement is HL-conditional at u0 = 2 and unlocated in print at any u0. No exponent moved. | 0, D | [recon-0830-rec-killrun.md](history/staging/recon-0830-rec-killrun.md) |\n| `Q-recon-0830-smooth-aps` | PARTIAL | Does any theorem on y-friable integers in arithmetic progressions (Granville, Fouvry-Tenenbaum, Soundararajan, Harper, Drappeau, Drappeau-Granville-Shao) supply the uniform o(1) equidistribution that attack-0830-varE-identification.md section 4 leaves open, for moduli up to y^(4/5) with the lam1 weight and the squarefree restriction? | NONE APPLIES AS STATED, and the step stays open; lim Var/E = 0.45546 stays HEURISTIC. Read at the page (eleven sources, arXiv text layers and the Acta Math. page image): every pointwise asymptotic is O(log q / log y)-precise (Granville 1993 Thm 1) or hypothesises log x / log q -> infinity, and every beyond-square-root level (Fouvry-Tenenbaum 1996, Drappeau 2015, Drappeau-Granville-Shao 2017, Pascadi 2023/2025) hypothesises y <= x^delta, i.e. u -> infinity, against our u in (1.2, 2]. NEAREST: Harper arXiv:1208.5992 (2012, a preprint with no journal version found) Theorem 2, the Barban-Davenport-Halberstam form, whose range (log^K x <= y <= x, Q <= Psi/log^A x, bounded u in the ineffective form) covers d up to L^(1/2)/log and whose weight (1 on all friable n, no squarefree restriction, no max over x' <= x) is unmet. The record's quantifier is also refined: the cell needs the lam0-weighted AVERAGE over d, not uniformity. PROVEN given the weighted form (H_w) of that theorem, derivation mine and unreviewed, with its two identities checked numerically in the producer: every two-branch cell of Xmix, both mirror cases, all bands, balanced blocks included, is o(ln^2 y), so the Bettin-Chandee kernel separation is not on this route; the reduction of (H_w) to Harper's theorem is OUTLINED and not written, and the three-branch type (measured -0.0002 at x = 19, trivial bound O(ln^3 y)) is untouched. Not TPC-strength. | 9 | [recon-0830-smooth-aps.md](history/staging/recon-0830-smooth-aps.md) |\n| `Q-recon-0904-sifting-limit-floor` | PARTIAL | What is actually known, in the owning convention and at the page, about LOWER bounds on sifting limits at dimension kappa > 1, and is the class-blind cap at beta_2 = 4.26645 a barrier or a method artefact? | Both. The corpus's [ABSENT] is wrong for lower bounds and right for extremal examples: Selberg's Lectures section 17 and Brady's 2017 Corollaries 1 and 3 ARE lower bounds on beta(kappa) at kappa > 1, invisible because Selberg's convention is the reciprocal a_k = 1/beta_kappa so his lower bounds are titled \"upper bounds for sifting limits\", while no kappa > 1 extremal example exists and Halberstam's own 2003 review says so; the best floor on beta(2) resting on a citation is 1.8394 from Brady's reduction evaluated at kappa = 2, unconditional and below the band, and beta(2) >= 2 follows from monotonicity by a padding argument DERIVED here and probably folklore; the cap at 4.26645 is a METHOD ARTEFACT at rung MEASURED, since the level-D LP is exactly the axiom-only information class and its calibrated floor of 3.3152 sits about 18 per cent below the DHR value and passed, 0 of 15, its first external test against the published beta_3 and beta_4; the band (2, 4.26645] is untouched by anything proven. | 0 | [recon-0904-sifting-limit-floor.md](history/staging/recon-0904-sifting-limit-floor.md) |\n| `Q-reconcile-bind` | ANSWERED | Can the unbound script tails be reconciled against their code and then bound by embed.js? | Six producers came back CLEAN-BOUND and one REBOUND with two documented label supersessions and identical results; nothing was DEFERRED and nothing DRIFTED outside the one pre-registered band move, fdecay-band-01's point 35.5 to 33.9 and band [30.8, 42.9] to [29.6, 38.3]. | none | [reconcile-bind.md](history/staging/reconcile-bind.md) |\n| `Q-record-deficit` | MIXED (lit-kourbatov-shortfall.md: ANSWERED; record-location-null.md: PARTIAL) | Does the 6.0 percent record-location deficit survive the corrected null, and what is it? | Survives (z = -3.3, five window cuts); Z5's premises on 'location' and on n_eff were wrong; corrected-null formula in §4. | Z5 | [lit-kourbatov-shortfall.md](history/staging/lit-kourbatov-shortfall.md), [record-location-null.md](history/staging/record-location-null.md) |\n| `Q-record-mechanism-0830` | PARTIAL | Which of three candidate mechanisms (sub-Poisson gap dispersion, the tile's exact gap law, Kourbatov's k = 1 conspiracy heuristic transported to k = 2) carries the residual 0.78 to 0.97 of Kourbatov's b, at what derived size, and what fraction of b is left? | The residual is carried by the twin-prime gap law AT HEIGHT and by nothing else tried: a new sieve to 1e11 measures that law under-dispersed (CV^2 rising 0.7276 to 0.9295 over seven decades) with a far tail steeper than exponential (log-slope 1.0638 at the top decade), and fed into the record null it carries 95.7 to 99.8 percent of the residual (d b_z = 1.0370 and 1.0806 +- 0.0014 against 1.0833, MEASURED on a model, paired, 2000 replicates); the tile's period-wide law (d b_z = 8.10) and Kourbatov-Wolf's k = 1 bin-count heuristic (4.20) overshoot by 4 to 7x and are closed as posed; the transport of a bare CV^2 is family-dependent (0.62 against 1.06 at CV^2 = 0.93), so \"computable from Var/E\" is false as stated; Kourbatov-Wolf's second-order scale a_c/abar tracks the measured CV^2 to within 0.04 at every decade with no parameter and is an ansatz, not a derivation; the shape statistic moves a third of the way toward the data (z_D from -1.64 to -0.97); what is left with no derivation is the gap law itself, and the top height band's b (0.937 measured against 1.69 to 1.98 predicted on 17 records) is the open falsifier. | Z5 | [attack-0830-record-mechanism.md](history/staging/attack-0830-record-mechanism.md) |\n| `Q-record-null2` | ANSWERED | How much of Kourbatov's shortfall coefficient b does an inhomogeneous-intensity record null carry, how much does a 6Z-latticed null carry, and what is the residual? | The inhomogeneous null carries essentially none of b (d b_z = -0.0012 +- 0.0001, MEASURED); the 6Z lattice's share is LAW-DEPENDENT, +0.1113 +- 0.0003 under the memoryless rounding-up law against -0.0060 +- 0.0003 under rounding to nearest, so the null side is a bracket of 16.1 to 25.1 percent of b and three quarters or more stays a residual with no mechanism; record-location-null.md section 8's argued 1.7e-4 lattice bound is wrong but no single number replaces it; the 15 percent estimator disagreement is pinned as an exact weighting identity; and the shape assumption behind every b reading is measured at 1.6 sigma and stays open. | Z5 | [measure-record-null2-0829.md](history/staging/measure-record-null2-0829.md) |\n| `Q-records-placement` | ANSWERED | Do the seven unswept twin-gap records 76-82, above 2^53, land uniformly inside their stretches, or is there square-anchor coupling in record-start placement? | Sealed alone before the producer existed: the uniform band, the fixed definitions, the registered readings, the calibration abort gate and the riders are all fixed here and no placement fraction, q or stretch boundary was evaluated; scored in records-placement-01.md. | Z6 (retired) | [records-placement-01.md](history/staging/records-placement-01.md), [records-placement-02.md](history/staging/records-placement-02.md), [records-placement-prereg.md](history/staging/records-placement-prereg.md) |\n| `Q-redteam-0818` | ANSWERED | Do the six headlines of 2026-08-18 survive being checked by someone other than their author? | Four CONFIRMED, two WEAKENED, one REFUTED: the second instrument's independence claim is refuted, the growth law weakens to G2 = x ln^{2+o(1)} x with the o(1) positive, and B is polylog, Brady's Problem 3 and DP1's consumer stand. | none | [redteam-2026-08-18.md](history/staging/redteam-2026-08-18.md) |\n| `Q-redteam-0820-empirical` | ANSWERED | Do the two held empirical headlines, perfold-error-model and item-x-offset, survive an adversarial pass before integration? | Both SURVIVE: every decisive statistic reproduced digit-for-digit on independent code, the blind window was re-sieved and re-scored 33/37 with independently re-derived bands, and the cofactor purity claim is certified; two sentences are weakened (the anchor-direction identical is only bounded at 6-10 per cent, and the @31 sigma calibrated claim outruns its controls) and one sentence is refuted as stated. | none | [redteam-0820-empirical.md](history/staging/redteam-0820-empirical.md) |\n| `Q-redteam-0820-math` | ANSWERED | Do fekete-1d, smoothness-front and import-hypergraph survive an adversarial pass before integration? | Mostly CONFIRMED, with one REFUTED as stated (import-hypergraph's \"the engine can never carry a ln-power\" is false for bounded-size edges, though the row's verdict survives for FGKMT-sized edges) and one WEAKENED (the freeze survives, the frozen value does not, under the trusted ladder). | none | [redteam-0820-math.md](history/staging/redteam-0820-math.md) |\n| `Q-redteam-0820-night` | ANSWERED | Do the night's two staged headlines, mp-derivation.md and rho-maximal-law.md, survive an adversarial re-derivation before integration? | Both survive: every decisive number reproduced on independent code, the fresh window re-sieved with a different engine reproducing all 37 X, the prereg custody at commit 303711b confirmed with the anchor genuinely fresh, and all 37 sealed (mu, lam) values and all 74 band integers reproduced exactly. | none | [redteam-0820-night-empirical.md](history/staging/redteam-0820-night-empirical.md) |\n| `Q-redteam-0820-night-proofs` | ANSWERED | Do the two proof-heavy held headlines of the 0820 night, anchored-01 and hsub-01, survive independent re-derivation? | Both stand: zero refutations and zero corrections to any quoted number, with every digit re-derived from the staircase-note's own lemmas or recomputed from the definitions in independent code; the doubling inequality itself was not attempted and remains posed, open and trap-free. | none | [redteam-0820-night-proofs.md](history/staging/redteam-0820-night-proofs.md) |\n| `Q-redteam-0820-structural` | ANSWERED | Do the Mirror-Sweep Lemma and its palindrome, and the advmin@11 = 16 headline, survive an adversarial replay? | Both CONFIRMED on independent re-derivation with zero shared code, advmin@11 = 16 re-proven end to end so that 16 < 34 kills every class-uniform cap; the correction that matters is one quantifier made explicit in the inversion sentence, since a family whose conclusion reads the chosen classes is not capped by 16 at the anchored point, and two precision defects sit in a Thorne citation. | none | [redteam-0820-structural.md](history/staging/redteam-0820-structural.md) |\n| `Q-redteam-0821-exponent` | ANSWERED | Do attack-rhoms-01's exponent claim and attack-doubling-01's certificates survive independent re-derivation? | No numerical discrepancy anywhere in either target, every re-derived figure reproducing to the printed digit; the 3.5+o(1) cap survives the attack it was most likely to die to but is not an unconditional bound on G2, and two wordings that run ahead of their custody grades are WEAKENED. | none | [redteam-0821-exponent.md](history/staging/redteam-0821-exponent.md) |\n| `Q-redteam-0821-structure` | ANSWERED | Do the W1/HL identity and the anchored-02 ladder survive re-derivation from the definitions on independent code? | Both headlines survive with zero refutations and all ten sub-claims CONFIRMED, carrying two recorded corrections: the -0.33 se figure's sign is presentation-dependent, becoming +0.32 se when duplicate-theta folds are collapsed, and the shadow guarantee is stronger than sold, a one-line theorem rather than a 76/76 empirical. | none | [redteam-0821-structure.md](history/staging/redteam-0821-structure.md) |\n| `Q-redteam-0821-wave2` | ANSWERED | Do the three held analytic attacks, b2mean, f4weak and quartic, survive an adversarial re-derivation? | All three survive with every load-bearing number re-derived independently, including four full N_k curves on a different inclusion-exclusion implementation and a c1 column from a freshly fetched OEIS b-file; two sentences are weakened, the cert/C2 widening being non-monotone, and no claim is refuted. | none | [redteam-0821-wave2-analytic.md](history/staging/redteam-0821-wave2-analytic.md), [redteam-0821-wave2-structural.md](history/staging/redteam-0821-wave2-structural.md) |\n| `Q-redteam-0828-census` | ANSWERED | Do destroyer-census-01.md, stretch-01.md and records-placement-01.md survive an independent re-derivation of every load-bearing claim, and may they leave HELD? | All three survive on arithmetic, every decisive number reproducing exactly on an independent engine; one sealed reading (records READ-3) is REFUTED as vacuous and four sentences need correcting before integration. | Z0, Z4, Z6 (retired) | [redteam-0828-census.md](history/staging/redteam-0828-census.md) |\n| `Q-redteam-0828-closures` | ANSWERED | Do the five 2026-08-28 closure notes stand, and do they stand for the reasons they give? | Three stand on corrected sentences (1d, Z5b, 0); one stands in half (10, the 1.074 counter-claim does not); one does not stand for its stated reason (1c, closed on a narrower window than the prior work's significant one, and its CLOSED has overwritten a PARTIAL in the generated index). | 10 (retired), 1c (retired), 1d, Z5b (retired), 0 | [redteam-0828-closures.md](history/staging/redteam-0828-closures.md) |\n| `Q-redteam-0828-engine` | ANSWERED | Do the five 2026-08-28 certificate-engine notes survive an adversarial re-derivation, and does the one-class freshness factor correction hold before it reaches a live document? | The one-class factor holds and is stronger than either note states; four numbers are refuted (first s>=10.82 is x=263 not 239, first non-empty kappa=1 level is @37 not @53, rho*g_max is off by one gap, the @17 certified share and per-term constant carry the wrong convention) and one proof step needs the max-over-y form of BV; the cluster's shared verdict stands. | 11 (retired), 8 | [redteam-0828-engine.md](history/staging/redteam-0828-engine.md) |\n| `Q-redteam-0828-head` | ANSWERED | Do the three 2026-08-28 head-residual notes survive an adversarial re-derivation of every load-bearing claim on independent code? | Mostly yes on the numbers and no on two framings: \"h/R -> 1 iff beta CV^2 -> 2\" is REFUTED (the limit needs only boundedness, the product controls the RATE), and 9 to 59 percent of the quoted h - R is the choice of comparator, since R is the continuum functional and a discrete uniform origin sees R + 1/2 exactly. | Z4 | [redteam-0828-head.md](history/staging/redteam-0828-head.md) |\n| `Q-redteam-0828-litimports` | ANSWERED | Do the six notes of 2026-08-28 survive an adversarial pass, and which of their load-bearing claims are wrong? | Every measured number reproduces on independent code; Saffari-Vaughan Theorem 10's range exponent is 6/11 not 1/11, which voids import-fracparts's one banked THEOREM at fixed M and corrects row 15 twice; four further weakenings, one refuted verdict line, no note refuted whole. | Z5, 4 (retired) | [redteam-0828-litimports.md](history/staging/redteam-0828-litimports.md) |\n| `Q-redteam-0828-quadpoint` | ANSWERED | Do the four HELD quadpoint notes of 2026-08-22 survive an adversarial pass: the capture identity's proof line by line, the depth law and its asymptote, the sealed prereg scoring, and the prior-art citations at page level? | The identity is PROVEN once its window convention is written down and holds at every K of 50 anchors under an independent implementation; four load-bearing sentences are wrong (section 4's gloss on Sum capU, v1 section 2's limit, the prior-art note's 2%, and the note's unconditional y* sentences), and none of the four defects opens or closes a route. | Z0, Z2 | [redteam-0828-quadpoint.md](history/staging/redteam-0828-quadpoint.md) |\n| `Q-redteam-0828-varE` | ANSWERED | Does the 2026-08-28 chain from the exact comb variance to lim Var/E = Pr[GD(2) > 2] = 0.45546 survive an adversarial re-derivation, and does the refutation of the 0.611 reading hold? | The constant survives at HEURISTIC and the 0.611 refutation is CONFIRMED and strengthened (the frozen out-of-sample half of the protocol also fails on the control); the correction is that TWO steps are open, not one, and that the theta=1 branch is an exact identity with a read theorem rather than a second-hand numerical match. | 9 | [redteam-0828-varE.md](history/staging/redteam-0828-varE.md) |\n| `Q-redteam-0829-measure-a` | ANSWERED | Do the three held measurement notes of 2026-08-29 (g2z2, g2-generic, provenance/x37) survive an independent re-derivation of every load-bearing claim, a custody and pre-registration audit, and a label audit against the wrong-direction scheme? | 96 load-bearing claims checked, 88 CONFIRMED on independent code or by hand, 7 WEAKENED, 1 REFUTED; the one refutation is a label, not a result: the g2z2 note's M2 and its producer's printed block both call 4.177e-1 the summed Poisson probability of a two-pair zone over p >= 3, while the code sums every decade including p = 2, whose single contribution is 2.6586e-1, so the true p >= 3 figure is 1.518e-1 and the MISS verdict is unchanged; no pre-registration in any of the three shows any sign of a prediction edited after its run, and birth-time evidence puts the note file before the producer file in two of the three, with the generic producer's inode anomalous and its claimed timestamps uncorroborated. | none | [redteam-0829-measure-a.md](history/staging/redteam-0829-measure-a.md) |\n| `Q-redteam-0829-measure-b` | ANSWERED | Do the three 2026-08-29 HELD measurement notes (measure-roughpair-null-0829, measure-record-null2-0829, zone-tail-02-0829) survive independent re-derivation of their load-bearing claims, their custody and pre-registration integrity, their (i)/(ii)/(iii) labels, and their proposed edits to other documents? | 59 load-bearing claims checked, 44 CONFIRMED, 13 WEAKENED, 2 REFUTED, no data defect found and all three producers bit-honest under embed --check; the worst finding is custody clerical rather than numerical, measure-roughpair-null-0829.md:161 quoting code-sha256 4813584b against the producer's banner 5caaa5173b4c1561, a hand-transcribed hash that does not bind the artefact it names; the second is zone-tail-02's scope claim that every head bootstrap in the corpus is three times too narrow, REFUTED as written because destroyer-census-01.md publishes no bootstrap at all and head-residual-factor.js bootstraps over gaps rather than zones, while the defect itself is CONFIRMED on an independent engine (1,910 distinct a_first among 17,700 top-band zones reproduced exactly) and is LARGER than stated, a cluster bootstrap giving an inflation factor of 4.10 against the note's implied 3.04; measure-record-null2's exact cov identity reproduces to 1.4e-15 from the OEIS ladder and its lattice arithmetic reproduces at every row, but its headline \"wrong by about three orders in b units\" is unit-inconsistent and contradicts its own section 3e's \"about one order below the effect\"; no pre-registration shows any sign of post-run editing and none of the three can be shown NOT to have been, because all three preregs live inside note files whose mtimes necessarily postdate their producers'. | none | [redteam-0829-measure-b.md](history/staging/redteam-0829-measure-b.md) |\n| `Q-redteam-0829-measure-c` | ANSWERED | Do the load-bearing claims of measure-tail-deficit-0829.md, lit-scourfield-2008.md, recon-0829-farfields2.md and recon-0829-escapes.md survive independent re-derivation at the record, and are their custody, prereg, provenance and (i)/(ii)/(iii) labels intact? | No verdict of the four notes is overturned; of 61 load-bearing claims 38 CONFIRMED, 14 WEAKENED, 6 REFUTED, 3 UNVERIFIED. Worst: lit-scourfield-2008.md line 79 quotes lit-smooth-divisors.md section 6 as saying the convention has \"no asymptotic machinery in this range at all\", a string absent from the repository and contradicted by that section's own words, and uses it to justify a correction; measure-tail-deficit-0829.md section 2 describes a superseded embed and does not disclose the --force the header records, although embed.js --check passes on all three hashes; and note 1's reported M-P1 failure at 19 -> 23 is a single-draw artefact, reversed here on 40 draws (shuffled 296.1 sd 29.1, Markov 264.0 sd 18.4, share 34.9 percent inside the registered band). Note 2's precision correction inverts a quantifier, the open step needing o(1/ln y) against Scourfield's O(1/log x); note 3's corpus does hold the per-prime eigenvalue at lit-pdf-holt-rudd.md line 259; note 3's section 3c re-prices a family already resident as the Vaaler completion of its own Lemma V remainder. Every verdict, and most of the arithmetic, otherwise reproduces on an independently written engine. | none | [redteam-0829-measure-c.md](history/staging/redteam-0829-measure-c.md) |\n| `Q-redteam-0829-objects-gb` | ANSWERED | Does every load-bearing claim of the two 2026-08-29 object reads (G2 and the bridge) survive independent re-derivation at the cited record? | 60 load-bearing claims re-derived at the record: 47 CONFIRMED, 8 WEAKENED, 1 REFUTED, 4 left UNCHECKED and marked so; the refutation is the G2 note's L5 formula, which pairs the corrected trusted zone K in [1.3946, 11.3568) with the superseded single-base expression (ln 66 + K)/ln 16, an expression that evaluates to 5.607 above beta2 at the top of its own stated range, and the three live-layer edits already applied by the orchestrator are correct as written. | none | [redteam-0829-objects-gb.md](history/staging/redteam-0829-objects-gb.md) |\n| `Q-redteam-0829-objects-zm` | ANSWERED | Do the load-bearing claims of z2-state-draft-0829.md and object-models-read-0829.md survive independent re-derivation at the record, and is the Z2 draft fit to become research/Z2-STATE.md? | 96 load-bearing claims re-derived at the record, 84 CONFIRMED and 11 WEAKENED and 1 REFUTED; the one refutation is a custody label, since the Z2 draft grades as unstamped hand arithmetic a table that sits in an embedded producer's OUTPUT block, and the draft is fit to become Z2-STATE only after eight named line fixes. | none | [redteam-0829-objects-zm.md](history/staging/redteam-0829-objects-zm.md) |\n| `Q-redteam-0829-theorem1` | ANSWERED | Is Theorem 1 of attack-AB-bounded.md (B <= 9 A(z)^2 (E(z)-1) = O((log z)^8)) a theorem, and at what rung does the corpus carry it? | PROVEN, and the scope now covers the whole chain: the L1 to L5 mean-square derivation and Theorem 1's bound on B are each re-derived here with no hypothesis, so _H <= 9 A(z)^2 (E(z)-1) H holds for every z, s and H as a mean over one period, the record's [INFERRED] is an under-grade, and the live layer's \"published as Opera de Cribro 6.18\" is wrong on two counts and must be replaced. | none | [redteam-0829-theorem1.md](history/staging/redteam-0829-theorem1.md) |\n| `Q-redteam-0830-doubling` | ANSWERED | Do attack-0829n-doubling-bridge.md and attack-0830-doubling-killrun.md survive an adversarial re-derivation on an engine sharing nothing with their producers, and do REFUTED rows 94 and 98 stand as worded? | The mathematics survives: the maxsum certificate Ghat(2s) <= maxsum_{K*+1}(T_s) and the product composition K*+1 <= prod(1+L_j) are each re-derived here and hold, the second at 21,641,346 nesting links over 6,012,804 killed runs with zero failures including the zero-kill fold, and every quoted figure reproduces digit for digit on a fresh engine (K*(16) = 17; G2(31#) = 348 @ 8813641451 x4 from both base tiles; sup msc 6.6364; the diagonal cells; the 2^N ratios 1.226/1.481/41.348). Row 94 STANDS and is if anything under-claimed, but its attribution is wrong in one direction: Lemma 1 at s = 128 alone clears the whole band, so the s = 16 walk corroborates rather than carries it. One statement is falsified: the bridge note's NOT-REACHED line puts 19#->37# out of reach, but column-major it returns here, reproducing the ladder row x = 37 and adding a FIFTEENTH step at s = 19, 20 (K* = 13, N = 4, C2 3.5200, certificate 3.8000), so the enumerable range ends at s = 20, not s = 18. Row 98 STANDS on its mathematics and is WEAKENED on one clause: \"the truth sits under it everywhere (sup w/a = 0.8295)\" attaches the CERTIFICATE's ratio to the word truth; the truth's sup is 0.6591. | D | [redteam-0830-doubling.md](history/staging/redteam-0830-doubling.md) |\n| `Q-redteam-0830-engine` | ANSWERED | Do attack-0830-buchstab-deep.md, attack-0830-comb-tail.md and attack-0830-anchored-ladder-17.md survive an adversarial re-derivation on code that shares nothing with their producers, and does REFUTED row 97 stand as worded? | Every headline number in the three notes reproduces exactly on independent code (26 of 26 comparisons, including F2/f2 = 3.6682 at sigma = 4.7088 by a near-analytic route, the 0-of-1512930 pair count, the tail defect at five levels, the full 419328-term Legendre split, and the whole @23 anchored ladder including K* = 27, floorU = 5364, truth 597475 and the minimal pool 1543 proven exact); one load-bearing DERIVATION is REFUTED — \"\\|B - 1\\| >= c forces sigma(q) < u_c - 1\" fails at @23 on 21134 (q, K) pairs for the engine's model B at c = 0.1 and on 33875 high-count pairs for the measured B at c = 0.01, because the step bounds every u in a weighted average by the largest one and treats the denominator as e^-gamma, so the \"level-free\" clause of buchstab-deep section 4 and REFUTED row 97's \"\\|B - 1\\| >= 0.01 forces sigma < 1.798\" must be re-graded from DERIVED to MEASURED-at-@23 (where the conclusion survives: the worst sigma is 2.159, still far below beta2 = 4.26645); row 97's \"the best sigma any level has\" is separately WEAKENED, since @29's head sigma is 5.5785 with width 1.5325 and @97's is 17.1422 with width 1.0000; comb-tail derives its main-term factor from prod (1 - 1/q_i) and measures it against prod (1 - 1/(q_i - 1)), which is what natal-cap-28 line 283 computes against its own line 20 and against certificate-engine.md section 1, a 0.9% difference at @23 that moves nothing certified; comb-tail also misquotes its own cited source as F1 in [1.5, 4] where that source says [1.5, 3] and says there is no bound at all at s < 1; four smaller slips are listed and none opens or closes a route. | 8, A | [redteam-0830-engine.md](history/staging/redteam-0830-engine.md) |\n| `Q-redteam-0830-fekete` | ANSWERED | Do the load-bearing claims of attack-0829n-hsubpow-K.md (the sign lemma, the K -> beta_bound map, the zone, the TODO.md base-16 correction) and measure-0830-delta-reader.md (the six pre-registered readers, their kill gates, the joint fit's calibration, the RB circularity argument) survive independent re-derivation, and are the two live-doc corrections they caused the right sentences? | Every number in both notes reproduces on independently written code (all 14 cells of the beta_bound / K_min table, all 48 cells of the reader-calibration table, all 24 figures of the real-control table, the margins, the sups, the 5.607 correction, the 10-of-81 count), and the sign lemma is re-proven as stated for continuous exact laws. Two claims are WEAKENED and one live-doc sentence needs a qualifier: the sign lemma's constant 1.0597 loses 43% of its value if the base floor moves from 2 to 3 and its whole shape fails for the stepped law that a primorial ladder must present, where beta re-enters the sup and a delta = 0 law already needs K = 1.37 at beta = 2, inside the trap; and TODO.md 1d's \"the sign of delta is unreadable for G2 at reach 79 by any instrument tried\" is wrong on the letter, since reader RB passed all four pre-registered criteria and read delta_hat = +0.80 +- 0.09, the note voiding it post-hoc on a circularity the pre-registration flagged but did not make a kill criterion; the honest sentence adds \"non-circular\". No claim in either note is refuted. | 1d | [redteam-0830-fekete.md](history/staging/redteam-0830-fekete.md) |\n| `Q-redteam-0830-floor-growth` | ANSWERED | Does the DERIVED growth half of attack-0830-rec-cheapest.md sec.4.2 (Omega >> z^{16s/9}/ln^8 z from four-prime Rosser exit chains) survive an adversarial re-derivation on independent code, does it really make REC(s, u0) false for every u0 < 16s/9, and does the blind slope test of blind-0830-omega-floor.md test the LAW or the CONSTRUCTION? | No break found in six attacks, so the growth half SURVIVES this pass and stays DERIVED (one failed break is not a proof, and the whole thing still rests on the exact half E1-E3 that this pass did not grade); but four claims are corrected - the threshold 16s/9 needs the unstated hypothesis s <= 3 (above it the four-prime LP maximum is 2(2s+2)/3, then 8, both below 16s/9, though the corollary that every legal u0 < beta_2 fails survives at every admissible s because 2 x LP-max > beta_2 for all s > 9beta_2/16 = 2.3999), the 2s - o(1) limit needs 3^k <= ln D / ln 8 so only k = 2 is available at every z below 1.1e9, A_i is >= not ~ the dyadic count, and the certified ratio does NOT rise monotonically (0.0118, 0.0128, 0.0101, 0.0057, 0.0051, 0.0093, 0.0336, 0.0801, 0.1658, 0.3545 on my own recount, which reproduces all ten S4b rows digit for digit); the blind test is ruled a test of the CONSTRUCTION's count and not of the law, since it cannot see E1-E3 or the CRT step; on the ruling Chris asked for, the floor does make the RML route a truth gap asymptotically and conditionally on E1-E4, and changes nothing computable, the literal construction first passing H = z^{u0} at z ~ 5.6e31 on the constant 2.93e-4 measured here; no exponent moved. | 0 | [redteam-0830-floor-growth.md](history/staging/redteam-0830-floor-growth.md) |\n| `Q-redteam-0830-floor-sign` | ANSWERED | Does the EXACT half of attack-0830-rec-cheapest.md sec.4 (E1-E4, the sign facts, the window bound T <= cc(r) + H - 1, the exact Omega, and sec.(d)/(e)'s calibration sentences) survive an adversarial re-derivation on independent code, and does it by itself constrain REC at any u0 without the growth law? | The mathematics STANDS as PROVEN and is stronger in two places than the note claims, but it constrains REC at NO u0 by itself: E3's identity is pure algebra and holds at every position (0 mismatches in 20.5M positions, both s), E2 holds at every divisor n > 1 under three disjoint routes at every s >= 1 and FAILS at s = 0.8, locating a hypothesis D >= z the note omits, E4 has 0 violations at every x over four complete periods with the wrap, and Omega is reproduced by an independent enumeration at all 26 tabled values and EXTENDED to z = 101 (285, 441, 684, 990, 1155 at z = 79, 83, 89, 97, 101), confirming blind-0830's exact 684 and 1155; three sentences are wrong -- E1's \"every pair with gcd \\| 2 is realised\" (exactly half are; the missing half is all those with 2 dividing one side), the ledger's and sec.0's \"251 against z^{u0} = 1.05e6\" (1.05e6 is HM, z^{u0} = 7.19e7, so the gap is understated 68.3x), and sec.4.4's reason for lam+(P(z)) = 0 (0.24 exceeds 1/8; the figure that closes the slab is 0.1199 at z = 89, under 1/8 by 4 percent) -- and sup\\|rho~\\| >= (Omega - M)/2 is loose: sup\\|R_1\\| = Omega + M exactly. Every exactly known log_z Omega is under 2 while REC needs it above u0 > 2, so the exact half touches REC at no u0 and every adverse consequence in the note still rests entirely on the growth half. No exponent moves. | 0 | [redteam-0830-floor-sign.md](history/staging/redteam-0830-floor-sign.md) |\n| `Q-redteam-0830-imports` | ANSWERED | Do the six lighter notes of 2026-08-30 (varE-identification, smooth-aps, rec-killrun, skeleton-door, at43-bigint, coherence) survive an adversarial pass on independent code and at the source pages, and which of their load-bearing claims are wrong? | No note is refuted whole and no exponent moves. Every decisive measurement reproduces on code sharing nothing with the producers: the varE mixed remainder and its 0.844 / 0.826 share below 2L and 94.4 / 98.7 percent unbalanced share at x = 13, 17; the Buchstab deviations and all three crossing levels log10 z* = 33.53, 68.80, 33.62; the skeleton door's 9.2 / -0.9 percent, G30_agg 0.1113 / 0.1011 and the 0.0102 cap at @13, @17, by a path that also verifies cap-36 Theorem A; the @43 parity exactness on all 26,157,448 scour classes with 0 above 2^54 and a constructed @47 counterexample; and the five recon-0828 verdict counts summing to 75, which REFUTES rec-killrun's brief-error clause and confirms the orchestrator's rider. Seven claims WEAKEN. The load-bearing one is smooth-aps's \"the weight is unmet in print\": Harper's J. London Math. Soc. 112 (2025) e70293 proves a Barban-Davenport-Halberstam asymptotic for an ARBITRARY sequence on exactly that dyadic all-classes object, and his own footnote says the missing max over x' can be incorporated, so the NEAREST citation changes even though the step still does not close (the 2025 range is sqrt(2x) < Q <= x and its Theorem 2 route bars sieved sets by name). Also: Henriot Corollary 2 applies, but to the ERRATUM's statement and under two hypotheses the note leaves unstated (eps < alpha/600, not \"every eps\"; F at prime powers); varE section 4's display (*) carries an unfilled placeholder and is vacuous by a factor L/d; the skeleton door's \"would tighten\" is wrong in sign at @17, where the closable block is negative and removing it RAISES G30_agg to 0.1020; Halasz-Montgomery is not applicable for the reason given (it is an abstract inner-product inequality); FGKT's heuristic quote is on p. 4, not p. 3; and the \"~13x had no source\" is a corpus miss, the retired G2-walk figure at CHANGELOG.md:5716. Not TPC-strength; nothing here opens a route. | 9, 0, D, 1 | [redteam-0830-imports.md](history/staging/redteam-0830-imports.md) |\n| `Q-redteam-0830-records` | ANSWERED | Do attack-0829n-parity-dstar.md (CLOSED, HELD) and attack-0830-record-mechanism.md (PARTIAL, HELD) survive an adversarial pass on independent code, does REFUTED row 95 stand as worded, and is TODO Z5's new title supported? | Both notes' measurements reproduce on independent code and neither headline is refuted: D* is confirmed EXACTLY, by a BigInt rational simplex written here, at 299 of 376 anchors with 0 contradictions (the other 77 exceed the exact solver's size cap), the old range's 172-figure gate is re-derived exactly at all 43 anchors with 0 disagreements, and an independent 30-wheel sieve to 1e11 reproduces note (b)'s seven decade gap counts, CV^2 and far-tail slopes to every printed digit, with the counts also matching the published pi_2(10^k). Four sentences are wrong or over-stated and are replaced here: REFUTED row 95, note (a)'s ledger, its section 6 and TODO Z5 all attribute the null's 0.025 agreement to the WIDTH law when 0.025 is the ln Q slope difference (the width-slope difference on the same 84 matched anchors is 0.031); note (a) section 3's \"the threshold sits in a gap of at least 0.25 at every anchor\" is scoped in its own producer to the true arm and the null arm's exact gap is 0.1212; P6's HIT rests entirely on the decade-2 collapse that section 5 discounts as a sampler artefact when it discounts kill clause (a); and note (b)'s \"d b_z = 1.0806 +- 0.0014\" carries only the Monte Carlo pairing error, while moving the far-tail slope by 0.02 (against a spread of 0.0174 across six fit windows on the same decade and 0.1196 across decades) moves d b_z from 0.9888 to 1.1720, and continuing the law at slope 1 above the last resolved point still gives 0.7752. REFUTED row 95's CLOSED verdict stands on kill clause (b), which is better powered than the note presents it (paired s.e. 0.0401 against a 0.15 band), and the row needs two clauses reworded; TODO Z5's title is supported for \"measured and not derived\", and the resolved part of the law carries 72 per cent of the residual on its own with the extrapolation carrying the rest. | Z2, Z5 | [redteam-0830-records.md](history/staging/redteam-0830-records.md) |\n| `Q-redteam-0830-rml` | ANSWERED | Do the three HELD notes of 2026-08-29 night and 2026-08-30 on the RML chain survive an adversarial pass — the six-arrow chain and the term-count cap lemma of attack-0829n-rml-proof.md, the same-object ruling and the applied replacement sentences of verify-0830-usup-convention.md, and the reflection identity, custody, pilot bounds and sealed forecast of measure-0830-rho-sup-z41.md? | The three load-bearing PROVEN claims survive independent re-derivation — the CAP lemma (every step, sine sum at worst ratio 0.585568, the constant 6, 0 per-modulus violations at z = 13, 17), the reflection identity rho~(W-3-y) = -rho~(y) (proved here and its two hypotheses verified at z = 13..37, residual <= 1.7e-10 at every position of z = 13, 17, 19), and the same-object ruling (an independent lattice reproduces the repository's term multiset exactly at z = 13..37 and its {Vabs > 0} set equals LSE's records at all ten levels); five of the seven cited exact sups and three of the five walked sup\\|R_H3\\| reproduce to <= 5.0e-7, and every figure of the sealed z = 41 forecast reproduces from the seven cited sups alone. Fourteen claims do not survive, none of them reopening a route. The load-bearing one is RML sec.2's ms_m/ask column: it measures the PROVEN mean-square BOUND sqrt(B H), not the ^{1/2} that REC names, the two differing by 36.53x to 200.45x on the tabled rows because saturates in H while B H does not, so sec.5's ordering (ask below the Gaussian factor at u0 = 3.0 and 3.5) is an artifact and reverses at every row against the true mean square. \"gauss -> 0\" (sec.5) and \"sup/rms <= sqrt(2 lnW) O(1) = z^{o(1)}\" (sec.3) are the same error twice, against the note's own sec.0: sqrt(2 ln W) = z^{1/2+o(1)}, measured gauss 0.7905 at z = 19 falling to 0.5645 at z = 1e7, limit 1/2. \"the recovery exponent is log_z(2 sup\\|rho~\\|/rms): about 0.99-1.29\" quotes the truth column, which is log_z(2 sup\\|rho~\\|) alone; the quantity named measures 0.5628 to 0.6712. \"REC would be moot rather than false\" should read false. \"u_cap -> 2s + o(1)\" and \"n ~ z^{2s}/polylog\" are asserted where only \"<=\" is proven and the data cannot decide: u_cap is 4.2652 at z = 47, crosses beta_2 only at z = 53, reaches 4.4891 at z = 79, and passes u_triv from z = 53. \"seven points spanning less than one octave\" is 1.509 octaves, still uncorrected in three places including the ledger verdict. verify-0830's applied \"polynomially many from z = 223 on\" is true at every z >= 13, and its z = 331 crossover silently sets A = 1 (A fitted to LSE's own Ssat moves it to 337). No exponent moves, no route opens or closes, and no ledger verdict changes. | 0 | [redteam-0830-rml.md](history/staging/redteam-0830-rml.md) |\n| `Q-redteam-0830-slack` | ANSWERED | Do attack-0830-head-remainder.md, attack-0830-tail-derivation.md and verify-0830-record-defects.md survive an adversarial re-derivation on independent code, and do the five riders the verification put on live notes stand as written? | Arithmetically they survive: every figure of the three notes that this pass could recompute reproduced to the printed digit on code written from the definitions, 0 assertion failures over 58 assertion call sites (most inside per-level loops), including every figure of the verification's three claims and the half-decade Delta_HL it quoted rather than recomputed. Four sentences do not survive. (1) verify-0830-record-defects.md:316, a replacement queued for head-residual-hl3.md sec.0, says the pooled value sits \"0.038 to 0.052 below the sub-window mean at all three decades\"; measured, the two lower decades run to 0.076 and 0.066, and the same note's own falsifier row says 0.038 to 0.076, so the wrong half is the one queued to land. (2) verify-0830-record-defects.md:379's un-measured \"+0.005\" residual inside the eighths is +0.0004 measured at sixteenths, so the de-pooled top-decade value is converged and the caution can be dropped. (3) the ruling that the class null is \"the matched figure\" for the tail is matched on residue class only: p'^2 is coprime to every q <= p', and the ensemble's rough-class offset is E_rc - R = 5.6714 -> 7.5357 over x = 7..29 against the class null's 5.6000 -> 6.0336, giving t/(R_shell + 7.5357) = 1.0123 against t/classNull = 1.0157. (4) attack-0830-head-remainder.md sec.3's top row is not gap-scale matched: E[g] 244.0 against 235.9, y_match capped at 19997, and on the note's own six ensemble points Delta_ens ~ E[g]^-0.312, so meas/ens reads 0.9640 not 0.9539 and the headline \"0.954 to 0.995\" reads 0.964 to 0.995. Three levels the notes recorded as out of reach are run here: the X2 group at x = 23, the tile identities and conditionals at x = 29, and the ensemble ladder at y = 29 (pipe/ens 0.9535, 40 s, against the head note's \"about fifteen minutes\"). No route opens or closes and nothing here touches Z2. | Z4, 9 | [redteam-0830-slack.md](history/staging/redteam-0830-slack.md) |\n| `Q-redteam-0830-zone` | ANSWERED | Do attack-0829n-X-upper.md, blind-0830-quadpoint-31607.md and blind-0830-quadpoint-c1c2.md survive an adversarial re-derivation and an independent re-implementation of their producers, and at what rung does the corpus carry each of their headline claims? | They survive on the numbers and mostly on the wording. On independent code every band sum, tier ratio, sealed forecast, band statistic and z of all three notes reproduces to the printed digit (X-upper 7.82 to 10.13 and the flat 378100.0; the fourth decade's 76 rows at 64 HIT and 12 MISS with the same twelve MISSes; the fifth decade's class A at both bands), and three attacks failed to break them (the paired-se attack on the D rows, a search for a fitted constant in C1, and gap-selection at the band edges). Four defects stand: the X-upper note's \"s_max 1.81 to 1.96\" is wrong in its own arrow convention, which means B3 to B8, since s_max peaks at B4 and reads 1.85 at B8; the c1c2 note's OLD R2 row at B14 is a boundary call that my 1-in-8 subsample scores the other way; the c1c2 note's custody line \"three tails, one code-sha256\" is a tautology of tailfmt.js and is not evidence of a seal; and the unscored y* check carries a Jensen bias of about +0.44 in y, which is the whole size of the C2 gap it is read as corroborating. Nothing here moves 4.2665 to 2, no rung rises, and all three notes stay HELD. | Z2 | [redteam-0830-zone.md](history/staging/redteam-0830-zone.md) |\n| `Q-redteam-0904-argmax` | ANSWERED | Does measure-0904-argmax.md survive an adversarial re-derivation of its positions, its congruence null, its residue chi-square and its own pre-registration scoring? | Every positional number in the note reproduces exactly on an independent fold-replication engine at x = 11..31, including the argmax lists, m(x), the top-K sizes and components, the per-prime pair counts, the tail counts and the F6 drops, so the note's data survive. What does not survive is three things and the largest is the null: two random TWIN SLOTS are congruent modulo W/p with probability (p-3)/(D-1), not 1/M, so the note's \"10^6 to 10^8 times the uniform null\" is inflated by 6.9 to 29.1 at these levels and should read 10^4 to 10^6, and the statistic is not a random variable at all but a deterministic function of each window (translate lemma, proven here and checked against every argmax pair count), which makes a null ratio a category error rather than a small correction; the note's own \"58 to 82 per cent are translates of another\" reports K - components when its wording needs K - isolated, which is 83 to 98 per cent; and the custody table in §2 quotes a code-sha256 (3019147f...) that the producer does not carry (d53a3fd1...), while §2's body and §5(5) still describe a widths annotation the rider says was already deleted. The scoring is faithful in verdict and wrong in arithmetic: F3 and F7 are scored at four levels, not five, so F7's falsifier fired at three scored levels and F3's tally is 2 of 4; the brief's \"three of eight\" does not exist on disk, both the ledger and §0 say four, and four is right. The exponent is untouched and nothing here is a route. | none | [redteam-0904-argmax.md](history/staging/redteam-0904-argmax.md) |\n| `Q-redteam-0904-floor-growth-2` | ANSWERED | Does the growth half of the adverse pointwise floor (Omega >> z^{16s/9}/ln^8 z from four-prime Rosser exit chains) survive a SECOND adversarial pass, and does a floor on the certificate at a CRT-planted window make REC(s,u0) false (a truth gap) or only unprovable by that certificate (a proof gap)? | SURVIVES, at the same rung and no higher: four new attacks and two unasked ones found no break, so the growth half stays DERIVED and red-teamed twice, which is two failed breaks on a derivation and not a proof; the proof-gap reading is REFUTED because REC is a statement about R_H, the remainder of the very certificate the floor bounds below, and the chain has no step where a certificate value must become a true remainder, so it is a truth gap for REC as the corpus states it; five corrections are owed to the record -- the REFUTED row is UNSCOPED and should read \"on the Rosser vector-sieve lattice at equal levels\" since only the Rosser instance is derived, the \"falsity beyond z ~ 5.6e31\" is an s = 2.698721 figure quoted without its s and the fitted crossing at the corpus's own measured family s = 3.0 is z ~ 10^15.3, the analytic p* window (4D^{1/27}, D^{1/9}/4) is neither necessary nor sufficient at reachable levels so the exact onset is 4.42e5 at s = 3.0 rather than the quoted 1.055e6, the four-prime construction has NO exit prime below z once s >= 7 so the first pass's \"8 beyond s = 5\" holds only on [5,7), and the first pass's escape-(a) density 1/W understates the planted class by the factor W/(n1n2) though the escape still fails by an exponential; the LP maximum 8s/9 is confirmed as an exact vertex and the corollary crossing reproduces 9beta2/16 to 4.4e-16, the pre-registered slope band held at a new s (top step 4.9025 at s = 3.0 against a model 4.8990), and no exponent moved. | 0 | [redteam-0904-floor-growth-2.md](history/staging/redteam-0904-floor-growth-2.md) |\n| `Q-redteam-0904-item0` | ANSWERED | Do the two HELD notes of 2026-09-04 on item 0, the second adversarial pass on the adverse floor's growth half and the L7 one-class-to-two-class transfer, survive an independent adversarial pass, and do the live-layer changes they propose survive verbatim? | Both notes SURVIVE at their own rungs and neither moves an exponent, but SIX of their proposed live-layer changes need rewording and two of their supporting derivations are wrong; every load-bearing number reproduces on independent code (18 of 18 A1A2 cells digit for digit, the LP maximum 8s/9 at 0 deviations of 16 with residual 4.441e-16, the crossing 2.399878284834 against 9beta2/16 at 1.097e-13, the fitted C = 1.1206e-4 and the s = 3.0 crossing 10^15.30, the onsets 4.390e5 / 1.111e6 / 1.860e6, ln(n1n2) = 7.4993e7, and the L7 CRT counts EQUAL at x = 7, 11, 13), so the truth-gap reading of REC is CONFIRMED and the growth half stays DERIVED and red-teamed twice, not proven; what does not survive is the scoping, the exit-window and the price wording: the proposed clause \"at equal levels D1 = D2 = z^s\" is NARROWER than the derivation supports, since over 1643 admissible upper-and-lower level splits the floor beats the decoupled window at every cell by at least 0.644444, so no level split escapes and the sibling rider \"asymmetric levels give 8s1/9 + 8s2/9 with s1 + s2 pinned\" is REFUTED as written (it holds only while both s_i <= 3); the record's z ~ 5.6e31 is u0-specific as well as s-specific (10^31.68 at u0 = 4.2165 against 10^35.36 at u0 = beta2, same s), so replacing it with a beta2-convention 10^15.30 mixes two conventions; the onset correction is right numerically and wrong in wording, since at the record's own s = 2.698721 the exact onset is 1.875e6, LATER than the quoted analytic 1.055e6, not earlier; the escape-(a) correction is right in substance but its \"(A1A2)^2 / (2 n1n2)\" is a LOWER bound on the split's contribution, and the two-sided version through lambda+(n) <= 2^omega(n) reads 10^{-3.004323e+7} against B ~ 10^{10.1}, still no contradiction; and on the L7 side the union-bound argument and the CRT count are CONFIRMED while the coupled vector-sieve price 2(1 + sqrt e) = 5.297442541400 is the FORCED-equal-level constant, the corpus's live figure being K_BF = 5.158064680330, reproduced here to twelve digits at (a, b) = (2.229949, 2.928115), and the sieve-on-holes is the vector sieve only after the Ford-Halberstam substitution step that sift-limit-attack.md 7b(1b) already owns; the \"no third mechanism excluded\" line is honest but incomplete, since weighted sieves with Chen switching are already priced and closed at sift-limit-attack.md 4.2; scope of death gains one row, L7's mechanism 2, which the growth note's own section 10 calls untouched while its sibling calls it dead; no exponent moved. | 0 | [redteam-0904-item0.md](history/staging/redteam-0904-item0.md) |\n| `Q-redteam-0904-r0-extension` | ANSWERED | Do the seven load-bearing claims of derive-0904-r0-extension.md survive an adversarial re-derivation against Tao's own text, and do its proposed live-layer changes survive? | Six of seven survive at their stated rung and one is refuted: Proposition A, the scope slip, row e'' and the equivariance transfer are CONFIRMED (the slip is strengthened, since Tao's own Example 2 is the target note's missing one-coordinate witness), Propositions B and D are WEAKENED (B's stated mechanism proves too much and its proposed insertion drops the qualifier that saves it; D's \"same trigger\" is near-definitional and its causal reading of the exponent band does not hold), and the claim that no corpus artifact certifies the window-count variance at width p'^2 is REFUTED by research/06-variance-theorem.js, which computes exactly that, brute-force verified at two levels; separately, EXEMPT throughout means \"Claim 1's H5 fails\", which is not the parity statement wall-note Face 1 is priced against, and the note's proposed live-layer text carries no H6 conditional. | Z2 | [redteam-0904-r0-extension.md](history/staging/redteam-0904-r0-extension.md) |\n| `Q-redteam-0904-sifting-limit` | ANSWERED | Do recon-0904-sifting-limit-floor.md's five claims survive adversarial re-derivation at the page, and in particular is the 4.26645 cap a method artefact at rung MEASURED? | Four of five claims survive with corrections and the headline does not. Claim d is REFUTED as stated on four grounds, each sufficient: the level-D LP fixes one omega-profile out of a class Ford defines by a ONE-SIDED inequality, and inside the classical (Omega_2(kappa,L)) budget a legal profile at x = 13 gives a calibrated reading of 5.0113, above beta_2 = 4.26645, which fires the target note's own pre-registered refutation of its floor; the calibration is a 1.72 multiplicative correction fitted at one point and it misses the second proven point beta(1/2) = 1 by 66 to 82 per cent, two and a half times the 28.7 per cent effect the headline rests on; the pooled 3.3152 averages four statistics of which only s* is the definitional analogue of beta, and that route reads 3.4678 here and 3.9487 at x = 43 in the corpus's own data, still climbing at +0.4297 per ln x with the fitted line reaching 4.26645 at x = 104; and 3.3152 is 22.2961 per cent below 4.26645 rather than the 18 per cent the note prints. Claim a is CONFIRMED with a second source off the OCR channel (Ford 2023 p. 37, \"Some authors, e.g. Selberg, refer to 1/beta(kappa) as the sieve limit\"), one page correction owed (Halberstam is p. 117, not p. 116). Claim b is CONFIRMED and strengthened: Brady's class is stated on his p. 10 and IS the axiom-only interval class Face 4 needs, his v_R table is reproduced here, and the identity v_{2d+2} = v_{2d+1} his proof asserts is confirmed at d = 0, 1, 2, so 1.819592 rests on proved facts alone. Claim c is CONFIRMED but demoted: beta(2) >= beta(1) = 2 is two lines from Ford's own one-sided (Omega) plus his own beta(1) >= 2, so the padding construction is correct and unnecessary. Of the eight proposed live-layer changes, two survive verbatim, five need rewording and one is refuted; the new SEARCH-CONVENTIONS row adds exactly three clearing phrases and changes nothing for any existing absence claim. | 0 | [redteam-0904-sifting-limit.md](history/staging/redteam-0904-sifting-limit.md) |\n| `Q-redteam-at37-score` | ANSWERED | Does the @37 scoring record survive an adversarial pass on its prereg re-score, its independence claim, its interpretation table and the cofactor extrapolation? | The empty-survivor scoring holds - every registered figure reproduces from the committed prereg text and the embedded census block, and the four skipped reference-level gates were run here and all pass - with two findings against the held report, neither touching the verdict: the one-level-drop figure is 8.2x rather than 4x, which strengthens the reversal claim, and the cofactor-one surplus sentence oversells. | none | [redteam-at37-score.md](history/staging/redteam-at37-score.md) |\n| `Q-redteam-zonegap-foundation` | ANSWERED | Do natal-onset-01, zonegap-01 and zonegap-prior-art, the three artifacts the Z2 pivot builds on, survive an adversarial pass? | The foundation HOLDS: all four natal-onset claims CONFIRMED, zonegap-01 confirmed at every reachable point by independent code using a different sieve and a different max-gap algorithm, and the prior-art spot checks all confirmed at page image; two prose defects found, both WEAKENED and neither refuted. | none | [redteam-zonegap-foundation.md](history/staging/redteam-zonegap-foundation.md) |\n| `Q-refuted-audit-0829-1` | ANSWERED | Do the closures recorded in REFUTED.md rows 1 to 17 survive re-derivation against the object definitions, the producers' embedded output and the corpus corrections through 2026-08-29? | Ten of the seventeen rows stand as written and seven are SOUND-NARROWER on wording that claims more than the record carries; no row is WEAKENED or UNSOUND, no route reopens, and the sharpest defect is row 16 crediting a published lemma whose own hypothesis excludes this corpus's operative range. | none | [refuted-audit-0829-1.md](history/staging/refuted-audit-0829-1.md) |\n| `Q-refuted-audit-0829-2` | ANSWERED | Do the closures recorded in REFUTED.md rows 17 to 34 survive re-derivation against the object definitions, the producers' embedded output, the 2026-08-29 corrections, and their own quantifiers? | All eighteen closures stand and none reopens; nine rows are SOUND and nine are SOUND-NARROWER, the worst being row 53's \"the search is finished\", which its own record contradicts in two places, and the others splitting into three classes: a mechanism clause that covers half its route (rows 47, 58), a universal quantifier wider than the sweep behind it (rows 49, 59), and a measured number carried without its calibration or its unit (rows 43, 46, 60). | none | [refuted-audit-0829-2.md](history/staging/refuted-audit-0829-2.md) |\n| `Q-refuted-audit-0829-3` | ANSWERED | Do the closure arguments behind REFUTED.md rows 34 to 51 (the greedy oracle through the bounded-differences family) hold when re-derived against the object definitions, the producers' embedded OUTPUT blocks, and the corpus corrections landed through 2026-08-29? | The band holds: 13 of 18 rows SOUND and 5 SOUND-NARROWER, no WEAKENED and no UNSOUND, so no route reopens; the worst finding is a factual inversion in the anchored-delta row, which asserts that delta exceeds the forced scale at 2 of 7 levels when the record's own table has \\|delta\\| < F at 7 of 7, and the same wording has already propagated into a second note. | none | [refuted-audit-0829-3.md](history/staging/refuted-audit-0829-3.md) |\n| `Q-refuted-audit-0829-4` | ANSWERED | Do the closures recorded in REFUTED.md rows 51 to 67 hold when the record is opened and the closing mechanism re-derived against today's corrected corpus? | All seventeen closures stand: 8 SOUND and 9 SOUND-NARROWER, zero WEAKENED and zero UNSOUND, no route reopens; the narrowings are wording that claims more than the record carries, worst of them row 67, whose closing clause quotes a section its own record headed POST HOC and carrying no verdict after the pre-registered primary went VOID on its matched control. | none | [refuted-audit-0829-4.md](history/staging/refuted-audit-0829-4.md) |\n| `Q-refuted-index` | SUPERSEDED | Where do recorded citations to the closed-route index resolve? | The current closed-route records are in OUTCOMES.md. This file is only a compatibility pointer for citations in recorded script output; it contains no separate outcome register. | none | [REFUTED.md](REFUTED.md) |\n| `Q-registry-glossary` | ANSWERED | What does each word of the programme mean, one term per object? | The vocabulary registry; adopted terms with dates, the anchored layer, and the calibrated entries (Var/E, hyperuniformity, the freshness dimension count) as corrected on 2026-08-28. | none | [GLOSSARY.md](GLOSSARY.md) |\n| `Q-registry-outcomes` | ANSWERED | What has each research approach established, where has it failed or stopped, and what evidence or changed input permits reuse or a further attempt? | One outcome register organized by research question and approach. Each record preserves its calibration, established result, limitation or failed step, evidence and reuse or revisit conditions. The closed-route table is part of this register. Git owns revision and authorship history; an open estimate is not a refuted claim. | none | [OUTCOMES.md](OUTCOMES.md) |\n| `Q-registry-research-readme` | ANSWERED | Which file in research/ answers which question, and in what order should a reader open them? | Router only: state lives in ../README.md section Status and G2-STATE.md section 0, the target in ZONE-POSTULATE.md, the queue in ../TODO.md; this file holds no mathematics and no summary of the programme. | none | [README.md](README.md) |\n| `Q-research-execution` | ANSWERED | How should agents use the integrated returns, avoid repeated work and select the next bounded research obligation? | The second dispatch from 1285d47 closed at 46fa948 and was independently reviewed on 2026-09-09. A/A2/B/C/D/E and V/W returns are present and integrated at their owning scopes. A2's corrected Type I estimate is accepted; D's fourth residual cut is retained and its remaining moment saving corrected to 7/200. C's low nonzero modes and E's aggregate scope are corrected. No new assignment is started by this integration; next obligations are candidates, not running work. Twin-prime infinitude and every sufficient signed margin remain OPEN. | C | [RESEARCH-EXECUTION.md](RESEARCH-EXECUTION.md) |\n| `Q-research-handoff` | ANSWERED | What must an incoming research agent understand, verify and return before its work can advance the current twin-prime argument? | Current handoff after the 2026-09-09 independent review of the dispatch ending at 46fa948. The full residual has four simultaneous complement conditions after D's regional estimate. A's centered truncation and A2's repaired low Type I estimate give the alternative S=C2*x+B+O_A(x/log^A x). B and all sufficient signed margins remain OPEN. The C3 absolute bound, family identities and scoped C/E failures retain their owning hypotheses. Execution is owned by RESEARCH-EXECUTION.md. | C | [RESEARCH-HANDOFF.md](RESEARCH-HANDOFF.md) |\n| `Q-research-round-validation` | ANSWERED | Which submitted results from lanes A, A2, B, C, D, E, V and W survive direct review, and what can be integrated or used to choose the next attempt? | Reviewed incoming commits 07ab47f through 46fa948 on 2026-09-09. A's truncation and A2's repaired low Type I estimate are accepted with bookkeeping corrections. B's ordinary-BV family identity survives; its uniform Dickman constant and complement condition are repaired. C's norm obstruction survives on all smooth inputs, but low nonzero Fourier modes are also 1-bounded. D's block estimate and fourth residual cut survive; its next moment saving is 7/200, not 17/200. E's rough-product distribution and aggregate constant 4 are accepted at the stated Wu uniformity reading; rational certificates close the two displayed tests for every fixed u>4. W remains OCR-only at the original 1974 theorem page. All sufficient twin-prime margins remain OPEN. | C | [research-round-validation.md](research-round-validation.md) |\n| `Q-residual-coverage` | ANSWERED | What portion of the full residual is controlled with uniform margins, and can the exceptional first branch improve that coverage rather than just a regional corner? | The first-branch and prime-free moments remove the exceptional norm condition. Full rectangles are controlled when delta+nu<19/25 or 5delta+2nu<123/50. Uniform twists justify exact coupled cuts: every fixed de<=x^theta with theta<19/25, and d^5*e^2<=x^lambda with lambda<123/50, as well as their union, have arbitrary fixed logarithmic savings. A concrete remaining endpoint sum has de>floor(x^(3/4)) and d^5*e^2>floor(x^(49/20)). Its twin margin is OPEN. The right zero-frequency budget, after an extra Cauchy inequality, limits the uniform product range; the left cross moment limits the other edge. Finite checks validate identities and masks, not asymptotic rates. | C | [residual-coverage.md](residual-coverage.md) |\n| `Q-review-0905` | ANSWERED | Does the present programme justify its latest stopping decision, and what is the next useful campaign toward infinitely many twin primes? | The computational checks pass and the DHR application survives this reading, but several summaries overstate the obstruction or misstate an implication. Recommend repairing those claims, retaining the fold framework, and testing it against Chen on long intervals before attempting a specifically stated parity-sensitive estimate. No proof of twin primes or improvement of an exponent is supplied. | none | [review-0905.md](history/staging/review-0905.md) |\n| `Q-review-request-0906` | ANSWERED | Which claims of the arithmetic campaign should an independent reviewer check, in what order, against which records, and what must the review return? | Completed review specification at the bounded scope recorded in handoff-review-0906.md and reports 20 and 21, with the subsequent round reviewed in round-review-0906.md. The dependency list records questions at dispatch, not current unfinished assignments. Imported proofs retain their review limits and the twin margin remains OPEN. Current assignments are in RESEARCH-EXECUTION.md. | C | [REVIEW-REQUEST.md](REVIEW-REQUEST.md) |\n| `Q-review-request-0907` | ANSWERED | Which claims of the 2026-09-07 session should an independent math agent check, in what order, against which records, and what must the review return? | Reviewed 2026-09-08 by an outside reader (history/reviews-0907/10): ranked claims 1 (conditional on the unread 1974 page), 2, 4, 5, 6 verified within stated scope; claim 3 verified with its quoted tolerances corrected (max deviation 1.8e-4, not 1e-4 or 1e-5); the Corollary 1 slip sharpened (fails for every representative other than 1, and mis-signs even then); ASSUMED items 4 and 6 discharged, 2 partially, 1, 3, 5 not. Twin-prime infinitude and every sufficient signed margin remain OPEN. Ranked below: the §6.2 demotion in the two-class manuscript (a change to a proof's dependency), the D_y identity reading (a change to what a census can show), the Erdős #687 attribution, and the two measurement artifacts. Twin-prime infinitude and every sufficient signed margin remain OPEN. | Z0 | [review-request-0907.md](review-request-0907.md) |\n| `Q-rho-maximal-law` | PARTIAL | What is the rho maximal law, what is its weakest sufficient form against the crossing, and where does it sit against print? | The pricing lemma turns any RML(alpha) with alpha < beta_2 = 4.26645 into G2(z#) << z^alpha ln^2 z, every step proven except the law itself; the sufficiency curve lambda_max falls from 2.31949 at z = 13 to 1.78446 at z = 47, the weakest measured point, and whether it ever reaches zero is model-dependent and undecided, so nothing here is an unconditional exponent. | 0 | [rho-exact-z31-01.md](history/staging/rho-exact-z31-01.md), [rho-maximal-law.md](history/staging/rho-maximal-law.md) |\n| `Q-rho-sup-z41-0830` | PARTIAL | What is the exact sup over all positions of the centred sawtooth remainder at z = 41 (period 7.42e12), and of R_H at H = z^3 and z^4, and what does the eighth exact point do to the truth's slope and to the u0 = 3 form of REC(s, u0)? | Not run in full; the exact z = 41 point prices at 4.8 to 5.4 h on this machine from measured per-position costs (naive 69.8, table 23.5, table-plus-windows 36.6 ns per worker; fleet 4.52 ns of wall at z = 41), above the 3.5 h rule, so what exists is a new engine (a PROVEN reflection that halves the walk, a P(29) periodic table worth 2.9x, per-block exact re-seeding) reproducing every cited exact sup at z = 13..37 digit for digit, sup\\|R_H\\| at H = z^3 walked for the first time at z = 19..37 (F3 = 26.4 to 35.9, the factor 2 was 1.5 to 2.2x loose), a sealed pre-registration (forecast sup 60.07, band [47.25, 76.37], kill needs sup\\|R_H3\\| >= 1807.417), one pilot twelfth giving lower bounds sup\\|rho~\\|(41) >= 70.651250 and sup\\|R_H3\\| >= 64.416936, and an eleven-segment plan for the box; no bound-and-prune exists on three routes; nothing asymptotic can move whatever the point turns out to be. | 0 | [measure-0830-rho-sup-z41.md](history/staging/measure-0830-rho-sup-z41.md) |\n| `Q-rho2-bound` | ANSWERED | Is there an analytic upper bound on (z) from its proven closed form, and does it matter for RML? | Already proven 2026-08-21 (attack-rhoms-01); the constant is now named, 170.88 z^{2s} ln^8 z; Chebyshev off any second moment diverges from RML (the true C_L = ms = 1 floor clears at z = 41 and 43 and misses by 2.155x at z = 47), so it is an ingredient, not a route. | 0 | [attack-rhoms-01.md](history/staging/attack-rhoms-01.md), [rho2-analytic-bound.md](history/staging/rho2-analytic-bound.md) |\n| `Q-rml-proof-0829n` | PARTIAL | Starting from Lemma V's PROVEN mean-square bound, what is the exact chain of implications to a two-class exponent below beta_2 = 4.26645, which single arrow is open, and does it close? | The chain from the proven mean square to an exponent below beta_2 has exactly one open arrow, the mean-square-to-sup recovery REC(s,u0): sup_x \\|R_H(x)\\| <= z^{u0/2 - eps} rms(R_H) at H = z^{u0} for one fixed u0 in (2, beta_2); every attempt here dies at a named place (term-by-term accounting is capped at 2s + o(1), the trivial-level exponent; position-union pays e^{theta(z)/q}; window smoothing and the P(y) lever move polylog only; the smooth-profile form is closed on hypothesis and priced above beta_2), the arrow is measured TRUE with certified margin z^1.72 to z^1.92 at u0 = 4 over z = 19..37 and the truth's slope 2.766 +/- 0.212 sits 5.8 se under u0 = 4 on seven points under one octave, so the failure is a proof gap at every computable z and undecided asymptotically; no exponent moved. | 0 | [attack-0829n-rml-proof.md](history/staging/attack-0829n-rml-proof.md) |\n| `Q-roughpair-error` | PARTIAL | How big is the empirical error of the rough-pair census, and is Z2 dead on the numbers? | Not dead on the numbers - the error sits below T at every one of 1,206 anchors, its scatter is sub-Poisson, its systematic part is 0.1% at the top band with the favourable sign and its exponent is 0.40 below the twin count's - but the same run moves the wall off the error term onto the depth, where the slack at the crossing is 1.056 and falling and the main term alone is 5.6 times the twin count at the sifting limit. | Z2 | [attack-roughpair-error.md](history/staging/attack-roughpair-error.md) |\n| `Q-roughpair-null` | ANSWERED | Does the independent-thinning null named in attack-roughpair-error.md's defects list return chi2/df = 1, and is the measured sub-Poisson dispersion of the rough-pair census error real or a normalisation artefact? | MEASURED, null-side, label (i), no bearing on the certificate: the independent-thinning null is binomial and returns 1 - p (0.8164 to 0.9808), not 1, so attack-roughpair-error.md's \"below 1 everywhere: sub-Poisson\" is WEAKENED and the registered F1 fires at u = 5; the deficit against the mean-matched null survives at pooled z = -5.87, -7.09, -11.39; and a CRT-exact null that only makes each prime's kill count exact in the window predicts 0.33 to 0.56, which the measured value EXCEEDS by 1.26 to 1.40, so D8 is interval-level in form, elementary in content, and not wall-relevant. | Z2 | [measure-roughpair-null-0829.md](history/staging/measure-roughpair-null-0829.md) |\n| `Q-round-review-0906` | ANSWERED | Which conclusions of the completed agent round survive review, and which corrections change the next research specification? | The fixed-band lift and continuous moving-window estimate survive the checked source interface. Interval stability additionally gives a dyadic scale-average logarithmic saving, so the blanket scale exclusion was too strong. The rate and missing prime-cofactor terms with s>1 or s'>1 still prevent a twin margin. Proper-prime-power branches already have a negligible bound on the fixed corner. The driver computes integer parity, not the CRT gcd partition; its total coefficient identities survive. Diagonal lower-bound, reachability and six-cut bookkeeping overclaims are corrected. No region, exact residual cut or twin lower bound changes. | C | [round-review-0906.md](round-review-0906.md) |\n| `Q-row11-closure` | CLOSED | Can the (1 - 1/e) greedy-coverage certificate of IMPORT-MAP row 11 ever fire? | No, and now on a producer rather than on prose: the map's paragraph-only pre-pricing reproduces exactly at all six figures it states and extends to the full ladder x = 5..79 with the certificate silent at every level under every constant that could be used, and the step from a density to an algorithm is proved in one line. | none | [row11-closure.md](history/staging/row11-closure.md) |\n| `Q-row11-closure-prereg` | OPEN | Does IMPORT-MAP row 11's (1 - 1/e) certificate reproduce, and does its test ever fire? | Pre-registration only, hand-written and committed before its producer existed: the claim being reproduced, what the producer must return, the pass criteria and the disclosed leaks are all fixed here, with the killing arithmetic still living only in prose. | none | [row11-closure-prereg.md](history/staging/row11-closure-prereg.md) |\n| `Q-row12-lonely-runner` | CLOSED | Does the lonely runner conjecture, or the Birkhoff and bounded-remainder branch, transfer to the Zone Postulate or to G2? | CLOSED WITH MECHANISM on two independent grounds, either sufficient: the general two-class problem is settled at the trivial bound, and the obstruction Tao names is our own configuration; the Birkhoff branch is subsumed and strictly worse from x = 13, and the row's payoff is a wall address rather than the published anchor the map priced. | none | [row12-recon.md](history/staging/row12-recon.md) |\n| `Q-row7-talagrand` | CLOSED | Does Talagrand's concentration on product spaces (IMPORT-MAP row 7) bound anything the corpus needs? | CLOSED WITH MECHANISM: the structural fit is better than the map wrote, since Theorem 8 needs no dependence structure and the shared-draw mechanism that closed rows 3, 4 and 10 does not close this one, but the per-coordinate effect c exceeds the largest tolerable value by a factor that DIVERGES (174x at @11 to 70,576x at @23), and the target hole is on the wrong side of the ensemble/anchor divide, so even a perfect bound would be inert. | none | [row7-recon.md](history/staging/row7-recon.md) |\n| `Q-scanstat-t37` | ANSWERED | What does the seventh exact level T37 do to the linear-in-ln D rule for H, and to the sqrt(m) refutation? | The sealed prediction H*5(T37) = 0.381254, band [0.369866, 0.392641], misses the measured 0.356548 +/- 0.0068 by 3.63 of its own s.e., so per the seal the linear form was a coincidence of a short ladder; the exponent series still rises but is concave in ln D, and G2(37#) is certified exhaustively over 217,929,355,875 slots. | none | [scanstat-t37.md](history/staging/scanstat-t37.md) |\n| `Q-scanstat-t37-prereg` | OPEN | What is the moving-sum exponent H at T_37, predicted before any producer for the pass existed? | Pre-registration only, sealed and committed alone: the prediction band, the criterion, the weak secondaries, the custody gates the producer must pass before reporting and the compute plan are fixed, and T_31 is explicitly excluded as another pre-registration's verdict. | none | [scanstat-t37-prereg.md](history/staging/scanstat-t37-prereg.md) |\n| `Q-scanstat2` | ANSWERED | Does the scan-statistic exponent law survive a sixth exact level, and is the tail factor the extremal index? | Both follow-ups come back negative: T_31 missed the pre-registered exponent rule from outside its own 95 per cent band, so the linear-in-ln D law describes five points and is not a law, and the tail factor is not the extremal index, its correction pointing the wrong way at seven of nine tested places; the sixth level confirms the kill on sqrt(m) harder, at H = 0.3460 +/- 0.0068 against 0.5. | none | [scanstat2.md](history/staging/scanstat2.md) |\n| `Q-scanstat2-prereg` | CLOSED | Does the linear-in-ln D exponent rule predict H(T31), and is the tail factor the extremal index? | Sealed and committed alone before any producer for the pass was written; scored in scanstat2.md, where both follow-ups come back negative: T31 missed the pre-registered band from outside, so the linear law describes five points and is not a law, and the tail factor is not the extremal index, with the correction pointing the wrong way at seven of nine. | none | [scanstat2-prereg.md](history/staging/scanstat2-prereg.md) |\n| `Q-search-conventions` | ANSWERED | In whose vocabulary do these objects live, and which prior-art searches have already been run in the owning convention? | The owning conventions are tabulated by object and the searches already run are listed so they are not repeated; the rule stands that a clean negative in our own wording proves nothing, and section 5 keeps what is still uncovered separate from what is covered. | none | [SEARCH-CONVENTIONS.md](SEARCH-CONVENTIONS.md) |\n| `Q-self-contradiction` | ANSWERED | Which documents in the corpus contradict themselves? | Twenty findings, two rated HIGH because they change what somebody does: theta-ladder.md's retracted crossing still asserted unmarked in three further sections, and G2-STATE.md's confirmed G2(41#) = 546 against three sections still carrying twelve terms; eight more change a number somebody quotes. | none | [audit-self-contradiction.md](history/staging/audit-self-contradiction.md) |\n| `Q-session-0828` | ANSWERED | What did the 2026-08-28 wave establish? | 21 agents, zero routes, four TODO items closed or answered, one HELD headline (Var/E = 0.45546, heuristic). | none | [session-0828-summary.md](history/staging/session-0828-summary.md) |\n| `Q-session-0829` | ANSWERED | What did the 2026-08-29 session (the Z/G object reads, the measurement wave, the red teams, the refuted-claims audit) establish, and what does it leave open? | Nothing moved an exponent or opened a route; the map moved: a Z2-STATE draft exists, both objects carry labelled legal-open-sets, the Gap Reformulation's discarded origin is proven not needed for the implication and measured worth one unit of exponent against a deficit of 2.2665, all 67 REFUTED rows re-derived with none reopening and 27 clauses corrected, and about forty live-layer and HELD-note defects were applied after verification. | none | [session-0829-summary.md](history/staging/session-0829-summary.md) |\n| `Q-session-0830-summary` | ANSWERED | What did the 2026-08-30 overnight campaign (rolling waves of up to ten agents on the queue rewritten after the 08-29 evening wave) change, and what is still open? | Neither exponent moved; twenty-one staging notes landed, all HELD; four routes closed (the skeleton door, the K*-product and fold-composition doubling bridges, the deep-ladder Buchstab transfer), TODO item 4 deleted and items D and 8 parked, the RML route carries an adverse derived floor whose growth law survived one sealed blind test and awaits a red team only Chris can call, Z2's fifth decade rejected the exact-Mertens main term for the omega-form crossing at 13 to 24 se, three record defects were verified on independent code and applied as riders, and a coherence pass corrected 27 lags in the live layer. | none | [session-0830-summary.md](history/staging/session-0830-summary.md) |\n| `Q-session-0904-summary` | ANSWERED | What did the 2026-09-04 wave (four Opus attacks on wall-facing questions, one orchestrator note, four Opus red teams, all verdicts applied) change, and what is still open? | Neither exponent moved; item 0's growth half survived a second adversarial pass so the route is a truth gap at rung derived-and-red-teamed-twice; Face 4's \"no lower bound on the kappa = 2 sifting limit is known\" was a convention failure and published floors at or below 2 are now cited, while the recon's headline that the cap is a method artefact was refuted by its red team; killer 2's coordinate is measured for the first time (argmax mirror-invariant, zero congruence pairs from x = 23, forced) after the multiplicity half turned out to be on the ladder already; no piece of R0 is both legal and parity-exempt with content; L7 is not a transfer; about thirty live-layer sentences corrected across seven files. | none | [session-0904-summary.md](history/staging/session-0904-summary.md) |\n| `Q-shadow-amplitude` | MIXED (shadow-amplitude-prereg.md: OPEN; shadow-amplitude.md: PARTIAL) | Where does the kill shadow's drift amplitude come from, and does the record's counting floor explain it? | The finite-y correction is an identity with no free parameter and both measurement routes agree 10 of 10, but the record's own counting floor is four to six times too small and its candidate explanation is the wrong half, so the label stays PARTIALLY EXPLAINED. | 5 (retired) | [shadow-amplitude-prereg.md](history/staging/shadow-amplitude-prereg.md), [shadow-amplitude.md](history/staging/shadow-amplitude.md) |\n| `Q-shadow-buchstab` | ANSWERED | Is the kill shadow the band-averaged pair-Buchstab integral? | SURVIVES WITH CORRECTIONS, and under the record's own pre-registration the verdict is SHAPE-ONLY rather than DERIVED, since y ~ 1000 misses D1 at 1.49x tolerance; the drift's amplitude is off by about 2.6x, the coefficient has the closed form 2 - 1/ln 2 = 0.5573049591 that the record missed, and anchored-windows section 5 reproduces 32 of 32 from an instrument sharing no code. | none | [shadow-buchstab.md](history/staging/shadow-buchstab.md) |\n| `Q-shadow-prereg` | ANSWERED | Is the kill shadow's 0.85 the band-average of the Unification-Law survival curve over the post-crystallization window? | Scored in shadow-buchstab.md: SHAPE-ONLY under the original y >= 997 rule; D3 passes in the reported ten clusters but D1 fails near y = 1000 at 1.49 times the 0.011 tolerance. Later finite-y normalisation and counting-floor corrections do not promote this verdict. The pair-Buchstab identification remains conjectural; the pre-registration body is preserved. | 5 (retired) | [shadow-prereg.md](history/staging/shadow-prereg.md) |\n| `Q-sharp-corner-transition` | ANSWERED | What is the energy of the complete sharp corner coefficient, and can the smoothed estimate be transferred through a negligible L2 transition? | DERIVED using named analytic inputs: for each fixed 00 both sharp squared norms and sharp-minus-smoothed squared norms are Theta_eta(x log^2 x) on the actual dyadic intervals. Thus an O_eta(x log x) sharp squared norm or negligible L2 transition fails in that range. Signed transition correlations and the global complement remain OPEN. No region, exact residual cut or twin margin changes. | C | [sharp-corner-transition.md](sharp-corner-transition.md) |\n| `Q-sharp-sieve-range` | CLOSED | Do the sharp sieve functions (Jurkat-Richert, DHR) make the two empty certificate-engine theorems non-empty at run levels? | NO at finite level: DH Thm 9.1 carries no written constant and the crude fundamental lemma is strictly better; as limit statements kappa=1 is first non-empty @37 (a 19.8-wide bracket; @53 is the first narrow one, 1.656) and kappa=2 @23. | 8 | [thm-sharp-sieve-range.md](history/staging/thm-sharp-sieve-range.md) |\n| `Q-shifted-prime-decomposition` | ANSWERED | Does decomposing Lambda(dk-2) give a provable saving on any part of the actual shifted-prime sum, and what exact arithmetic remains outside the imported hypotheses? | Both second Type I terms are O_H(x/log^H x) by classical Mobius BV. The residual is an explicit weighted sum over dk-ev=2, equivalently an average of two-linear-form Mobius correlations with growing coefficients and possibly one-point intervals. Its required one-sided improvement is OPEN. No twin lower bound or novelty is claimed. | C | [shifted-prime-decomposition.md](shifted-prime-decomposition.md) |\n| `Q-shifted-prime-mobius-sums` | ANSWERED | At reachable x, do the shifted-prime Mobius and Liouville sums, the cleanest instance of the parity input the twin reduction lacks, show any size or sign structure that a random-sign model does not, and how loose is the trivial bound on M(x) used for the centered tolerance? | MEASURED to x=2^38 (10.9e9 primes). All four sums sit at random-sign size: \\|U\\|/control rms at most 2.8 at any j, slopes 0.28 to 0.65 against the control's 0.43 to 0.47 with per-draw spread 0.19 to 0.30; the pre-registered structure falsifier did not fire. M(x)/x is at most 8.8e-5 in magnitude for j>=29 against the trivial bound 0.374, so the -4x/25 tolerance in moving-cutoff-parity is loose by four orders at these scales; a proof still needs a bound on M, which is the same parity object. The squarefree density of p+2 among primes is 0.747911 at 2^38 against 2*Artin = 0.7479116. pi(2^j) matches OEIS A007053 at every j. Nothing here bears on the asymptotic conjecture. | C | [shifted-prime-mobius-sums.md](shifted-prime-mobius-sums.md) |\n| `Q-sift-limit-discards` | PARTIAL | Where does the dimension-2 sieve discard our structure, and can any of it re-enter to beat beta2 = 4.2665? | Every asset checked against the five discard points has no known consumer and no entry point, and EH and GRH change nothing here; Lemma V's mean-square form is PROVED and hands back a sup bound, the strata are self-similar and it still does not close, L has no combination law of its own, and the 529 covering route is closed rather than unfinished. | none | [sift-limit-attack.md](sift-limit-attack.md) |\n| `Q-sigma31-calibration` | ANSWERED | Is the sigma behind the @31 X-channel detection calibrated at @31 itself? | Calibrated at its own level for the first time: seven known-truth control draws give sigma_true/sigma_reported = 1.127 at the point, 95 percent bracket [0.745, 2.293] (pooled with the earlier draws, [0.772, 1.664]), which restates the detection at z = -2.06 under the point calibration and closes both T2.b findings; the detection weakens rather than dies. | none | [attack-sigma31-01.md](history/staging/attack-sigma31-01.md) |\n| `Q-signed-divisor-grouping` | ANSWERED | Can grouping different fibers while retaining their Mobius signs control an explicit part of R, and what signed error remains after that grouping? | For every fixed H>0 the actual remainder restricted to de<=x^(7/10) is O_H(x/log^H x), although each sign's mass there is at least c*x*log(x) eventually. This follows from a written classical-input derivation, with finite algebra and archived-factor checks. The remaining de>x^(7/10) contribution reduces to an explicit weighted CRT endpoint discrepancy whose required one-sided improvement is OPEN. | C | [signed-divisor-grouping.md](signed-divisor-grouping.md) |\n| `Q-signed-moment` | PARTIAL | Is there an argument for the block that does not pay sqrt(M) at the first Cauchy inequality or that extracts cancellation from the R=0 class, and what are its budgets at the corner a=b=1 and at the target box (delta,nu)=(8/25,9/20)? | The true diagonal has an upper bound x^(1+o(1)) on the transition band, not a uniform nonzero lower bound. Lemma A controls only u1=u2,h1=h2; unequal proportional R=0 pairs remain outside it. Lemma B and the now-priced joint Cauchy arrangement add no region. The former general Holder floor is valid only under additional coefficient assumptions, not for arbitrary sparse coefficients. Conditional off-diagonal budgets and the random-matrix ceiling remain conditional and heuristic. No necessity of using both signs, full-corner estimate or sufficient twin margin is established. | C | [signed-moment.md](signed-moment.md) |\n| `Q-single-alignment` | CLOSED | Does the head's single-alignment fold recursion for M(x, x^3) give a proven-shaped handle on the per-fold multiplier that the max-over-alignments recursion lacked? | The answer splits: the multiplier IS different in shape, measured - exactly 1 at 225 of 240 folds from x = 53, total spend 1.09 nats against a tile-shaped ladder's 7.12, measured M(1613) = 2220 against a tile-shaped 3.7e5 - but the route to a growth law through the recursion closes anyway, because the growth is initiated by the window's expansion into fresh ground, a boundary term that is the localized gap problem re-posed and that no fold reaches. | 0d (retired) | [localized-single-alignment.md](localized-single-alignment.md) |\n| `Q-singleton-fiber-audit` | ANSWERED | Can the one-point fibers of the coupled remainder be discarded or controlled in absolute value at scale x, and what do the actual weights and archived intervals show? | The grouped coefficient is zero whenever either integer is prime. Each sign's ungrouped singleton mass is at least c*x*log(x) eventually for an absolute c>0, by a written derivation from named classical inputs. Finite full-interval and archived-prefix checks validate the partition. The signed singleton and longer-fiber contributions remain OPEN; no twin lower bound or novelty is claimed. | C | [singleton-fiber-audit.md](singleton-fiber-audit.md) |\n| `Q-skeleton-bound` | PARTIAL | Does the aggregate 30-skeleton bound hold, and can it be certified at every level? | Certified as an exact BigInt inequality at six levels @11 through @29, margins 0.287 to 0.406, with the Skeleton Collapse Theorem proven for all x and all q and an exact exception criterion; the all-x form is open and its blocking term is prime-equidistribution of window residues rather than branch algebra. | none | [natal-cap-30-skeleton-bound.md](natal-cap-30-skeleton-bound.md) |\n| `Q-skeleton-decide-0830` | ANSWERED | Should TODO item 4, \"Skeleton Equidistribution: restate or demote\", be restated around the modulus-W mass or demoted off the board, and what exactly is banked either way? | DEMOTE. Over every scour prime the branches on which the door is a fixed-modulus question carry 9.2, -0.9, 0.2 and 0.5 percent of the skeleton at @13, @17, @19, @23 (the -10.8 and 5.5 percent on record were 9 and 2 percent subsamples), so the door as named removes at most 0.0102 from a G30_agg whose open part is 0.094 to 0.126; on the open side the phase never wraps and no equidistribution statement remains, only the inequality; nothing on the live board consumes G30_agg < 1/2; no first move exists that is not already ANSWERED. Banked regardless: the Skeleton Collapse Theorem (PROVEN, all x, all q) and G30_agg < 1/2 CERTIFIED at six levels @11..@29. | 4 (retired) | [decide-0830-skeleton-door.md](history/staging/decide-0830-skeleton-door.md) |\n| `Q-skeleton-door` | ANSWERED | Is the Skeleton Equidistribution Conjecture the blocker on the anchored calm? | It is not the blocker as far as the door can be seen: Theorem A (Trapezoid Cancellation) is proven for all x, q and branches, @29 is certified as an exact integer inequality at G30_agg = 0.1176 with margin 0.3824 over 7,863 scour primes, and the decay-law shortcut is REFUTED, the six levels being flat within their own spread with every fit made strictly worse by adding @29. | 4 (retired) | [natal-cap-36-skeleton-door.md](natal-cap-36-skeleton-door.md) |\n| `Q-small-divisor-kernel` | ANSWERED | Does reciprocity separate the j<=x^(1/20) kernel of grouped-divisor-moment (21) at delta=8/25, nu=9/20, and do the Duke-Friedlander-Iwaniec / Bettin-Chandee and Kuznetsov interfaces then supply the required saving? | The separation is exact and cheap: reciprocity splits index by index, the correction phase has sup norm and total variation O(A/(MN))=O(1/x) at this box, and Mellin separation of the endpoint weight costs x^epsilon with L1 mass min(1,Ax/(MN)), so the whole block reduces to a pure trilinear Kloosterman fraction with unimodular twists. The interfaces then fail by explicit amounts: Bettin-Chandee Theorem 1 gives block exponent 129/125, above the required 1 by 4/125 and above the classical 103/100; DFI (1.1) gives 1267/1200; every box Bettin-Chandee controls is already controlled, verified on 19005 rational boxes; the audit's spectral diagnostic factor is trivial for all common divisors up to x^(19/50). Closing this box inside the same interface needs the (M+N) exponent below 27/140 or the (AMN) exponent below 9/28. Target (21) and the global twin margin remain OPEN; nothing is added to the controlled region. | C | [small-divisor-kernel.md](small-divisor-kernel.md) |\n| `Q-smooth-sieve-literature` | PARTIAL | Which nearby published results should be imported instead of rediscovered, and do their methods give a more focused next question for the complete global remainder? | Existing sources supply the finite-difference mechanism, one-point concentration and quadratic profile optimization; shifted divisor-sum estimates admit fixed shift two under their own support and smoothness conditions. A derived application of corrected Henriot bounds gives small-prime mass O_sigma(delta^sigma x)+O_epsilon(x^(39/40+epsilon)), uniformly in delta for fixed 0floor(x^(77/100)) as a fourth simultaneous condition of W_dagger through the Perron machinery, changing only the summation domain of E_dagger. The target box (407/400), the corner, the uniform product threshold 19/25, the sign of E_dagger and the global twin margin are unchanged; the margin remains OPEN. | C | [structured-dispersion-estimate.md](structured-dispersion-estimate.md) |\n| `Q-supported-coefficient-dickman` | ANSWERED | Can the whole supported coefficient be priced using existing literature before enumerating its factor cells, and does the resulting separate-sign sieve estimate supply a positive twin margin? | DERIVED from classical inputs: at delta=1/100 the harmonic limit is H_delta=D(25/12)-1+E_delta with 0<=E_delta<=24/22!, hence -13/2, where I is the explicit ordered two-factor integral below. A three-prime left input and rough composite right input suffice; prime partners are subtracted using an upper sieve. The witness survives every fixed delta<1/50 and every averaging profile with the same plateaus. Consequently the negative-only consumer is false, and liminf Phat/(C2*x)>1/2. The signed twin margin remains OPEN; literature novelty is unestablished. | C | [switching-negative-mass.md](switching-negative-mass.md) |\n| `Q-tail-deficit-model` | ANSWERED | Does the exact compound-geometric thinning of the previous level's own gap histogram, with the qualifying-gap constraint g = 0, +/-2 (mod q) carried through the ladder, predict the tile's tail-deficit factor G2/(mbar ln D) at zero parameters, and does that make object-models-read-0829's D5 residue-level classification constructive rather than inferred? | NO. The zero-parameter CRT-constrained thinning predicts a tail HEAVIER than the tile's at every level, delta_M2 = 0.6461, 0.6746, 0.7021 against a measured 0.4577, 0.4463, 0.4791, so the pre-registered \"classification was wrong\" arm fires and D5 returns to OPEN with C3's constructive route REFUTED; the constraint is informative but small, carrying 4.0 to 10.0 percent of ln(null/truth) at the far tail, or 7.7 to 14.4 percent read against a rate-matched control, against the grain's ORDER at 39.4 to 54.6 percent at the deepest abscissa each level resolves; at shallow abscissae the ordering reverses (B 8.7 against C 4.8 at x = 23, u = 4), which the exact operator on a shuffled true grain confirms without any model (maximum 276, 354, 426 against the true 204, 258, 348), and the order effect is now localised (sections 7 and 9): a first-order Markov word from the true adjacent-pair statistics carries 32.3, 34.2 and 60.8 percent of ln(shuffled/truth) at the deepest abscissa of the three folds, and a SECOND-order word from the true adjacent-triple statistics carries 97.3, 97.9 and 102.6 percent there, r = 0.972, 1.061 and 1.000 on draw-mean maxima over 40, 20 and 10 draws, so the folded tile's tail is MEASURED to be a function of the previous grain's three-consecutive-gap statistics and of essentially nothing longer-range; that surrogate is seeded on the true previous grain at one fold, is not iterated, predicts no delta and does not revive C3, and the 348.0 sd 0.0 at 29->31 rests on ten draws. | 0 | [measure-tail-deficit-0829.md](history/staging/measure-tail-deficit-0829.md) |\n| `Q-tail-derivation-0830` | PARTIAL | Does the tail's measured law c = 0.7522 ln^2(p'^2) and its surplus derive, and which part of the coefficient is the tile's own (no prime input) against which part needs Hardy-Littlewood? | PARTIAL, kill clause fired on the coefficient: the ensemble object (the same functional averaged over all p# translates) derives exactly, E_all = R + 5/2 and E_odd = R + 3 with R = Sum g^2/2W, computed exact at x = 7..29 with every conditional the rough-origin candidate needs, and its Mertens constant e^{2gamma}/(8 C2) = 0.6007 is 1/(2 C2) times rho(2) = e^{2gamma}/4 by construction; every route from there to the anchored 0.7574 passes through rho(2)'s limit (the sharp form of Assumption A, algebraically equivalent to HL) and through the zone's gap shape CV^2 -> 1 (an HL statement), so the anchored coefficient is HL in disguise and nothing beyond HL is derived; the tile's own (1 + CV^2)/2 is 0.6306 -> 0.7532 over x = 7..29 and not settled; the record's 0.7522 = HL x 0.9644 x 1.0298 is two non-HL-sized factors cancelling at this height; the rough-origin candidate has an exact ensemble counterpart E_rc/E_cls = 1.0050 -> 1.0523, right sign, wrong size and trend against the anchored 1.0157, not identified; one number derives that the record measured: the class-null offset E_cls - R = 5.600 -> 6.034 exact against the record's 6.05, replacing both the prereg's 0.5 and the note's 7.5; and the corpus's \"R + 1/2\" is the other convention, under the tail's own the constants are 5/2 and 3. | Z4 | [attack-0830-tail-derivation.md](history/staging/attack-0830-tail-derivation.md) |\n| `Q-tail-maximal` | CLOSED | Is a tail-only maximal inequality a smaller ask than the full one, as attack-AB-bounded section 4.4 supposed? | No: the split is exact but is a reparametrisation rather than a localisation, the tail is maximal and saturates the Gaussian law, a fixed cut buys nothing asymptotically, and the tail is not a smaller ask but the whole ask. | none | [attack-tail-maximal.md](history/staging/attack-tail-maximal.md) |\n| `Q-tailcount-transport` | ANSWERED | Does the Tail-Count Transport survive an independent adversarial re-derivation? | The inequality, the certificates and the NO-CHAIN verdict all STAND on an implementation written from the operator definition, with a seventh fold added exact (31: 348 = 348); three corrections follow, the stated equivalence is false at fold 29, zero loss is exactness-by-relaxation rather than a tail fact, and the measured spends are 1, 1, 2, (1 or 2), 3, 2. | none | [verify-tailcount-transport.md](history/staging/verify-tailcount-transport.md) |\n| `Q-tau-repricing` | CLOSED | Does the tau(m) re-pricing of the x-free bound stand? | The separability claim is true and re-derived four ways, but the re-pricing substitutes one quantity for a different one: summed over the modulus set the tau(m) bound IS the divisor-pair count up to the vector sieve's fixed factor 3, and used as a bound it is 4 to 5x weaker than the exact quantity in the Fourier basis and 23 to 44x weaker in the divisor-pair basis. | none | [attack-tau-repricing.md](history/staging/attack-tau-repricing.md) |\n| `Q-the-dials` | ANSWERED | Which variables can this programme actually push, and which is Chris missing? | The square frontier constrains certification from roughness alone, not every arithmetic method. The current campaign uses long intervals, a classical Chen benchmark and an OPEN signed estimate; uniform maximum-gap control is a sufficient secondary target. Finite slopes do not determine an asymptotic exponent. | none | [THE-DIALS.md](THE-DIALS.md) |\n| `Q-the-lens` | ANSWERED | Why is the tile the right object, and what does the lens buy? | Completeness is PROVEN and elementary (every twin prime above p is a twin slot of T_p) and the recursion is exact; nothing here is claimed as new mathematics, and the triage rule of section 5 states what the lens does not buy. | none | [THE-LENS.md](THE-LENS.md) |\n| `Q-theta-ladder` | CLOSED | Does the all-positions exponent theta turn over below 2, converge to exactly 2, or settle above 2? | The answer is above 2 with an amendment, and the route is retired regardless: theta does not turn over (it dips and comes back, need/z^2 = 0.3550 to 0.6119 at z = 13..31 and rising from 31), convergence to exactly 2 is not supported (e = 1 rejected at 95% in all four fits), point estimates run 2.37 to 2.99 without settling, and OUTCOMES.md records the certificate route as RETIRED because the sharp maximal law it needs is itself TPC-implying. | 0 | [theta-ladder.md](theta-ladder.md) |\n| `Q-theta-last-gap` | CLOSED | Can the last 0.209 of theta_total be recovered by sharper absolute-value accounting? | No: theta_total stays 1 and the obstruction now carries an exponent, the absolute-value ceiling is c*H^theta_total with c > 0 rather than H*polylog, so the shortfall is a power H^0.212 and the entire budget any constant could ever buy is O(lnln H / ln H) against a break-even of 1.208983. | none | [attack-theta-last-gap.md](history/staging/attack-theta-last-gap.md) |\n| `Q-theta-selfconsistent` | ANSWERED | What does the exact-supremum theta ladder give when re-run self-consistently in H? | All five published self-consistent values reproduce exactly from two independent implementations; z = 31 is now measured at nP = 588, nP/z^2 = 0.6119, so the published 1.1776 is overstated by 1.93x, continuing the pattern, and at z = 47 a prefix walk finds the certificate failing at H = 1.0412 z^2, so the self-consistent unconditional requirement leaves the zone budget by z = 47. | none | [theta-selfconsistent.md](history/staging/theta-selfconsistent.md) |\n| `Q-transition-energy-review` | ANSWERED | Does the sharp-energy chain in sharp-corner-transition.md survive an adversarial check, and do the upper norm, the small-fixed-eta lower norm and the failed negligible transition each stand at their stated scope? | All ten assigned rows survive at their stated scope; no counterexample or decisive contradiction was found. The upper norm, the small-fixed-eta lower norm and the refuted negligible L2 transition each stand separately. Two repairs are citation-level only: the integer Barban-Vehov input is Graham 1978 quoted by Chen An, and the note should record why the k=Q constant is absolute. The de la Breteche-Dress-Tenenbaum PDF was refetched and its SHA-256 matches the pinned hash; its (1.5) and Theorem 1.1 supply exactly the imported (3) and (4). eta_0 remains positive and unextracted, and the signed sums, E_out and the twin margin remain OPEN. | C | [transition-energy-review.md](transition-energy-review.md) |\n| `Q-transition-energy-review-spec` | ANSWERED | What independent checks are required before the sharp-energy and non-negligible-transition conclusions can guide the next signed research attempt? | Bounded review assignment with explicit source, algebra, interval, error and quantifier checks; per-claim verdicts and dependency consequences are required. The assignment is specified, not executed. | C | [transition-energy-review-spec.md](transition-energy-review-spec.md) |\n| `Q-transition-joint-budget` | PARTIAL | Can the mixed and smoothed terms be handled jointly with the transition pair, and what complete inequality would make such an estimate useful for the twin consumer? | The exact split, cutoff-average identities and separate upper budgets survive. Averaging cutoff norms after triangle and Cauchy is saturated for sufficiently small fixed eta, but signed cutoff arguments and joint cancellation are not closed. R_00 need not be estimated separately under every grouping. The short-window constraint is corrected with a bounded majorant and a safely reduced lower constant. No consumer input or signed improvement is supplied; E_out and the sufficient twin margin remain OPEN. | C | [transition-joint-budget.md](transition-joint-budget.md) |\n| `Q-transition-joint-budget-spec` | ANSWERED | Can the mixed and smoothed terms be handled jointly with the transition pair, and what complete inequality would make such an estimate useful for the twin consumer? | Bounded assignment to audit the exact four-term decomposition, price a concrete joint treatment, and return a sufficient one-sided or common-scale inequality with all open terms explicit. The decomposition and existing norm budgets are inputs, not new findings. | C | [transition-joint-budget-spec.md](transition-joint-budget-spec.md) |\n| `Q-transition-research-execution` | ANSWERED | How should the core research handler execute and integrate the three bounded transition assignments while retaining the central signed estimate? | Prepared execution contract with a fixed mathematical baseline, three independently assignable specs, handler-owned signed research, dependency checks, shared-file ownership, reporting requirements and stopping rules. Execution is complete; the corrected results are in transition-round-review.md and transition-round-audit.md. The specification itself supplies no arithmetic estimate. | C | [TRANSITION-RESEARCH-EXECUTION.md](TRANSITION-RESEARCH-EXECUTION.md) |\n| `Q-transition-round-audit` | ANSWERED | Which conclusions of the completed transition round survive direct review, which require correction, and what research decisions remain justified? | The sharp upper norms and sufficiently-small-fixed-eta lower norms survive. The integration contained substantive errors: an omitted lower cutoff, an incorrect Mellin formula for the transition, a conflation of singleton divisor fibers with long cofactor fibers, an unpriced source representation and overbroad closures of joint or cutoff arguments. Owning notes and current records are corrected. No signed improvement or sufficient twin margin is established; the corrected cofactor and joint-correlation interfaces remain open. | C | [transition-round-audit.md](transition-round-audit.md) |\n| `Q-transition-round-review` | ANSWERED | What survives the executed transition round after direct review, and what combined estimate is still missing? | The energy chain survives at its original scope; the integration's missing cutoff, Mellin representation, fiber identification, normalization and general closure claims are corrected. The exact cofactor and joint-cutoff formulations remain useful. No signed estimate, controlled region, exact cut or sufficient twin margin changed. Only the stated norm-only procedure and the arbitrary-factor fixed-shift relaxation are refuted at their specified scopes. | C | [transition-round-review.md](transition-round-review.md) |\n| `Q-transition-signed-estimate` | PARTIAL | Can the signed transition kernel (E1) be estimated beyond its absolute norm budget, and if not, what exact input does the consumer need? | No signed improvement. Derived: the proper-prime-power part of R_11 is O(x^(39/40+epsilon)); the prime part is an exact beta-weighted sum over pairs of linear forms of two-point Mobius correlations with coefficients up to x^(6/25+2eta) and x^(1/20+2eta); the Cauchy budget is at most 2B*eta*x*log^2(x)(1+o(1)) and the fiber triangle budget 4*eta^4*x*log^4(x)(1+o(1)), so every existing inequality loses one of the two signs. The cutoff-difference identity converts T at pn into a sharp-window coefficient times log(p)/L plus two boundary windows; the entropy-decrement route then needs bounded coefficients, a short-interval one-point input for T and a log-weighted consumer, none available. The required one-sided input is stated exactly in section 5. The twin margin remains OPEN. | C | [transition-signed-estimate.md](transition-signed-estimate.md) |\n| `Q-transition-source-match` | ANSWERED | Do the inspected divisor-correlation, shifted-convolution and averaged-Chowla statements directly supply a usable estimate for the exact transition kernel? | No direct supplier was established among the inspected statements. The cofactor exchange and Theta identity are exact, but their scalar coefficient mass is not a full bounded-class representation for T. The transition Mellin kernel requires a sharp subtraction with a 1/s tail. Fixed-divisor singleton fibers differ from potentially long cofactor fibers. The fixed-shift arbitrary-factor relaxation is refuted; broader signed and cutoff methods remain open. | C | [transition-source-match.md](transition-source-match.md) |\n| `Q-transition-source-match-spec` | ANSWERED | What bounded source investigation could supply a usable signed estimate for the transition kernel, and how must its hypotheses and losses be priced? | Specification for up to three detailed source-interface assessments, including coefficient representation, scale quantifiers, normalization, separation errors and exact downstream payoff. No new theorem match or literature-absence result is claimed. | C | [transition-source-match-spec.md](transition-source-match-spec.md) |\n| `Q-twin-reduction` | ANSWERED | What has the arithmetic campaign of 2026-09-05 and 2026-09-06 actually established about the twin sum, at what calibration, what exactly remains, and what would suffice? | Consolidates the checked regional reduction S=C2*x+E_dagger+O_H(x/log^H x), its sufficient unbounded-scale margins, coefficient definitions and source dependencies. Later complete smooth and centered formulations are owned by RESEARCH-HANDOFF.md and their linked derivations. No new arithmetic estimate is supplied. Imported theorem proofs retain their stated review limits; every sufficient twin margin remains OPEN. | C | [TWIN-REDUCTION.md](TWIN-REDUCTION.md) |\n| `Q-two-class-lower-bound` | PARTIAL | What does the two-class LOWER bound on G2 say, and does it conflict with attack E's measured law? | The finite range cannot decide and it is not close: with both constants set to 1 the transferred bound and attack E's measured law do not cross until x = 10^7327 while the exact ladder stops at 79, so neither result refutes the other and no computation ever will. | none | [attack-lower-bound.md](history/staging/attack-lower-bound.md) |\n| `Q-two-class-lower-bounds` | PARTIAL | How large can G2(x#) be FORCED to be? | The best PROVEN lower bound costs nothing, the Kalmynin-Konyagin substitution is carried out here (DERIVED, independently checked, not refereed), and the certificate ladder is MEASURED with an oracle test whose load-bearing caveat is stated on its own; against the target x'^2 and the proven x^(4.2665+eps) upper bound the question stays open. | none | [two-class-lower-bounds.md](two-class-lower-bounds.md) |\n| `Q-two-moire` | ANSWERED | Do the Grain and the Scour, generated by disjoint prime alphabets, force twins to be infinite? | The provable content is exact CRT factorization: over the JOINT period the miss count is guaranteed positive, so aggregate alignment is arithmetically impossible, but that is an aggregate statement over a period vastly wider than the zone and does not reach TPC; the nearest unclaimed theorem is a two-class Jacobsthal upper bound at ANY exponent, and exponent < 2 would prove TPC outright. | none | [two-moire-argument.md](two-moire-argument.md) |\n| `Q-u-frame` | PARTIAL | Does restating the Zone Postulate as one inequality in u, stepping on odd numbers rather than on primes, buy anything against the wall? | It buys form and closes off a class of dead ends, but no difficulty and no slack: the requirement ln c(p) <= 2 ln p/p is exactly what the per-prime analysis already gave and the measured ladders run at about the full rate, so this is a change of coordinates rather than of terrain. | none | [U-FRAME.md](U-FRAME.md) |\n| `Q-u2-engine-depth` | ANSWERED | Is the certificate engine's operative depth heading to u = 2, and does that make TODO item 8's payout form TPC-strength? | It is not: the measured operative depth runs u = 4.191 at @17 down to 3.557 at @97 and is still falling, so the brief's conditional does not fire and item 8 is shown neither TPC-strength nor safe; a WALL-ADDRESS in the weakest sense, every figure SCRATCHPAD-GRADE. | 8 | [u2-engine-depth.md](history/staging/u2-engine-depth.md) |\n| `Q-unreached-sources` | ANSWERED | What is actually in the three sources SEARCH-CONVENTIONS section 5 lists as unreached? | The 2009 SeqFan thread on A144311 does not exist, now a read negative over 19,964 archived messages with zero hits; Paseman's n is the number of distinct prime factors and at that reading his shape is asymptotically weaker than our x^{4.2665}; MathSciNet stays paywalled and unswept. | none | [audit-unreached-sources.md](history/staging/audit-unreached-sources.md) |\n| `Q-usup-convention-0830` | ANSWERED | Are lemmaV-sup-extension.md's u_sup and u_sat and attack-0829n-rml-proof.md sec.4.1's CAP(z) statements about the same object (level D = z^s at s = 3.0, same moduli, weights and normalisation), so that the sec.4.1 correction in passing applies, or does a level-convention mismatch void it? | SAME object, VERIFIED at the code: both notes build the lattice through buildTerms(z, z^3), the modulus sets and Vabs(e) agree to 6.9e-18 at z = 13..23 and the count convention matches at all ten levels, the cited Ssat and u_sat reproduce on the RML lattice to every printed digit at z = 13..19 and the full-level alternative does not (19.544 against 19.602 at z = 13, 45.827 against 50.314 at z = 17); so the PROVEN cap Ssat <= CAP <= z^{2s+o(1)} applies and refutes lemmaV-sup-extension.md's asymptotic prose at lines 485-486 and 508-516 (2^pi(z) moduli, a C^pi(z) theorem as the reachable end), while sec.4.1's attribution sentence overreaches by calling that note's \"reading\" a shape it lists as one of two indistinguishable fits and flags as its likeliest error; neither ledger verdict line changes. | 0 | [verify-0830-usup-convention.md](history/staging/verify-0830-usup-convention.md) |\n| `Q-var41` | OPEN | What does the stable law predict for Var(41), and what can the tenth Var/E point pin? | Pre-registration only, sealed and committed alone before any Var(41) engine exists: it freezes the prediction, a band taken from the law's own residuals at z <= 37, the derived z(41) prediction, and the honest statement that one more point cannot separate a limit from a drift. | 2, 9 | [var41-prereg.md](history/staging/var41-prereg.md) |\n| `Q-varE-identification-0830` | PARTIAL | Can either open step behind lim Var/E = 0.45546 (the identification delta*(X - X_dec) -> 0, or the theta = 2 mean-coefficient replacement) be proven, and if not, which single inequality does not close? | Neither closes, and the two named steps are one statement (Conjecture 1 of variance-note sec.10; the model half is PROVEN in varE-limit-theorem.md). The remainder is re-split exactly at five levels: the corpus's X2 column doubles the positive half of one shift pattern (X2c = 2.66 against the group sum 0.09 at x = 19, PROVEN by rebuild), so its Xmix is overstated 2.3x; the true mixed remainder is MEASURED at -0.49 ln y and 82 % of it sits BELOW 2L, in the two-branch type, in the unbalanced range min(d,e) <= L^(2/5) (99.7 % of that cell at x = 19), outside the range of every bilinear Kloosterman-fraction bound; the divisor-distribution estimate the record names (n > 2L only) is therefore not the whole open step. Henriot's Corollary 2 (read at the page, erratum read) applies as stated and cuts the moduli to n <= L ln^(2+o(1)) y, PROVEN given that theorem; the one inequality left is the uniform o(1) equidistribution of y-friable squarefree integers in progressions to moduli up to y^(4/5), NOT SEARCHED at the page. The limit 0.45546 stays HEURISTIC. Not TPC-strength. | 9 | [attack-0830-varE-identification.md](history/staging/attack-0830-varE-identification.md) |\n| `Q-varE-limit` | PARTIAL | Does lim Var/E on the diagonal window exist, and what is it? | The theta=2 mean-coefficient step is neither proven nor refuted: the replacement error is exactly a sum over shifts of W(h) - V(h), it splits into a c=0 group (needs no decoupling) and CRT-mixed lags (open); measured ratios true/model 1.0098, 1.0039, 1.0014, 1.0013 at x = 13..23 with delta(X - X_dec) ln y falling rather than settling, so the error is O(1/ln y) or better MEASURED on four levels that exclude growth and nothing finer, and the limit 0.45546 stays HEURISTIC with varE-spectral's second step, its own limit theorem, still open. | 9 | [lit-dickman-variance.md](history/staging/lit-dickman-variance.md), [lit-smooth-divisors.md](history/staging/lit-smooth-divisors.md), [varE-asymptotic.md](history/staging/varE-asymptotic.md), [varE-exact-ladder-01.md](history/staging/varE-exact-ladder-01.md), [varE-limit-theorem.md](history/staging/varE-limit-theorem.md), [varE-spectral.md](history/staging/varE-spectral.md), [varE-theta2-proof.md](history/staging/varE-theta2-proof.md), [varE-theta2-step.md](history/staging/varE-theta2-step.md) |\n| `Q-vc-prior-art` | ANSWERED | Does arXiv:2208.06442 contain or overlap the VC-dimension = 4 finding of import-vc-nets? | DISJOINT at theorem level; not found there nor in the reachable one-hop neighbourhood; Helmbold-Sloan-Warmuth 1992 Thm 3.1 owed (abstract only). | none | [lit-vc-multiples.md](history/staging/lit-vc-multiples.md) |\n| `Q-verify-audit-numbers` | ANSWERED | Is audit-numbers.js circular? | Not circular in the way that would matter most, since it reads no other repo script, but 15 of its 117 checks certify nothing and 17 more are artifact echoes among transcribed constants; the one bug class the charter was written against had nothing in sections A to W to fire against and still has nothing, and one from-scratch re-implementation found a check depending on an unnamed hard-coded phase choice. | none | [verify-the-verifier-numbers.md](history/staging/verify-the-verifier-numbers.md) |\n| `Q-verify-cofactor-convolution` | ANSWERED | Does the cofactor-convolution identity at the anchor hold, and does it test the joint law? | RESTATED with corrections: the arithmetic is right, an independent re-derivation reproducing every figure to the digit at all five levels and confirming the asserted closed-form step, but the framing does not survive, since X is pinned by the identity X = M - Nbar + n_0 so the whole comparison collapses to one cell and the joint is measurably not a product elsewhere; the model error is 0.52 per cent, not 0.050. | X | [verify-cofactor-convolution.md](history/staging/verify-cofactor-convolution.md) |\n| `Q-verify-monotone-depth` | ANSWERED | Does the Monotone Depth obstruction, which retires dial 4, survive an adversarial re-derivation? | NARROWED: every measurement in the record reproduces, several to the last digit, but the mechanism as named is false as a general law and the corpus already holds a counterexample to it; what survives is an accounting of three channels priced in tenths of a nat against a need of 6.700, 8.191 and 9.637 nats, so dial 4 becomes priced slack with one channel open and unpriced. | 0e (retired) | [verify-monotone-depth.md](history/staging/verify-monotone-depth.md) |\n| `Q-verify-record-defects-0830` | ANSWERED | Do the three defect claims of 2026-08-30 (attack-0830-varE-identification.md sec.8 on the doubled X2 column; attack-0830-tail-derivation.md sec.7 on the corpus's \"R + 1/2\"; attack-0830-head-remainder.md sec.1 on the decade-pooled Delta = 0.6214) reproduce on independent code from the definitions, and which of the corrections they list apply? | Claim 1 CONFIRMED: the corpus delta*X2 column is the positive half of the p \\| h-2 pattern doubled, that pattern's mirror is the p \\| h+2 pattern (W-(-h) = W+(h) exactly, W-(h) != W-(-h) at 9 to 900,679 shifts), the group sum 2X2 reads -0.5603 .. 0.1846 at x = 7..19 against 5.7147 .. 5.3266, and Xmix is overstated 2.305x at x = 19 (6.9x at x = 7); X, X1 untouched. Claim 2 AMENDED: R + 5/2 and R + 3 are exact for the tail convention (all and odd origins, asserted at x = 7..23) and t/(R + 3) = 1.0228, but head-residual-factor.md:70 is the HEAD's forward convention where R + 1/2 and R + 1 are exact, so that correction does not apply; and p'^2 is 1 mod 6, where the tail constant is 5 (t/(R + 5) = 1.0181), with the class null (1.0157) the matched figure, so 1.0228 is one unmatched convention replacing another. Claim 3 CONFIRMED as an artefact, AMENDED on mechanism and on the half-decade reading: 0.6214 pooled against 0.6592, 0.6705, 0.6736 at 2, 4, 8 sub-windows; the between-slope (Simpson) piece the note derives is half of the pooling term (0.0187 of 0.0378 at halves, 0.0261 of 0.0522 at eighths), the other half is the within slope's own fall with height weighted by Var(g); and re-reading the halves from their eighths puts HL BELOW the measurement at 5 of 6 half-decades by 2.5 to 3.4 percent at the top decade, so \"within 1.6 s.e., sign alternating\" is itself a pooled statement. | Z4, 9 | [verify-0830-record-defects.md](history/staging/verify-0830-record-defects.md) |\n| `Q-verify-the-verifier` | ANSWERED | Does the QC gate's alarm ring when a defect of each real class is present, rather than merely reading green? | Of the 31 blind mutations the existing selftest covers zero: it is a reachability test proving each alarm can ring, never a sensitivity test asking how far a defect must move before the alarm stops, and its fourteen controls are all obvious silences; the ranked blind spots are listed with a proposed fix each, led by out-sha256 being written and never read. | none | [verify-the-verifier-checks.md](history/staging/verify-the-verifier-checks.md) |\n| `Q-w1-singular-series` | ANSWERED | What is the endpoint comb W1 in the literature's terms? | W1 is exactly the Hardy-Littlewood singular-series ratio for (0, 2, v, v+2) over the twin series squared, truncated to the folded primes and normalised by a v-independent constant, PROVEN algebraically and VERIFIED as exact BigInt-rational equality, so the fold-factor comb is the twin-twin correlation comb evaluated at v = theta_p. | none | [w1-singular-series.md](history/staging/w1-singular-series.md) |\n| `Q-width-repairs` | ANSWERED | Can the five loaded containers found by the width sweeps be guarded without invalidating any embedded figure? | Forward protection only: every repair is a widening or a bound check, storage-only by construction, no corpus figure was retroactively corrupt and none moved, with maxgap-law.js at 0 of 1389 figures unreproduced and attack-foldL-01-census.js byte-for-byte on the same out-sha256. | none | [width-repairs.md](history/staging/width-repairs.md) |\n| `Q-width-sweep` | ANSWERED | Which big-level enumeration and census scripts carry a fixed-width container that is safe at every level run and aliases silently above a threshold? | Three instances: two CORRUPT (the one-word residue mask that corrupted the census from x = 37 up, and a Uint16 index into the scour-prime array in the @37 census, both since fixed) and one that threw rather than aliased; every internal identity in the corrupted producer is a COUNT identity and stayed PASS throughout, because the alias corrupts which prime, not how many. | none | [width-sweep-census.md](history/staging/width-sweep-census.md) |\n| `Q-width-sweep-attacks` | ANSWERED | Does the attack and analysis script family carry the fixed-width-container defect class that corrupted the @37 census? | No live corruption across 123 scripts, every stored value derived at the level actually invoked and fitting its container; one hazard is serious, attack-x-offset-02-profile.js storing an index into the scour-prime array in a Uint16Array, the identical mechanism in the identical role as the census defect fixed at 605ce83, and it fires at @37. | none | [width-sweep-attacks.md](history/staging/width-sweep-attacks.md) |\n| `Q-wrap-identity` | ANSWERED | Can T4 be computed without enumerating quadruples? | Yes: the 4-point wrap identity exists and is verified through @13 against the certified value, turning @17's dead T4 (4.9e16 quadruples) into an 11-minute computation, but at a precision that certifies T4 itself and not yet the quartic bound. | none | [natal-cap-32-wrap-identity.md](natal-cap-32-wrap-identity.md) |\n| `Q-wrongdirection-audit` | ANSWERED | Are any of the ten live targets secretly TPC-strength, before the sessions are spent on them? | The audit closes no target: two of the ten come out TPC-strength and three more carry a TPC-strength face under a quantifier they are likely to drift into, which is a labelling result and not a reason to drop them; item 1e is settled only because the answer was already on disk, and one route-blocking number in it is hand arithmetic and unstamped. | 1e (retired) | [attack-wrongdirection-audit.md](history/staging/attack-wrongdirection-audit.md) |\n| `Q-xchan-at29-prereg` | OPEN | Does the joint-deficit closed form survive a blind test at @29? | Pre-registration only, committed alone before any producer existed: the statistic, the predictions adopted verbatim from the record, two acceptance bands, the validation gate the instrument must clear before any @29 number is reported, and what each verdict does to TODO item X. | X | [xchan-at29-prereg.md](history/staging/xchan-at29-prereg.md) |\n| `Q-xchan-at37-offset` | ANSWERED | Does any registered offset-correction candidate for the ~3.8 law survive at @37? | Sealed and committed alone before any @37 census of any kind existed, fixing the candidates, the sigma model and its projection band, the scoring rule and what each verdict does to item X's offset clause; scored in xchan-at37-score.md, where every registered candidate is killed and the number survives an independent recount. | X | [xchan-at37-offset-prereg.md](history/staging/xchan-at37-offset-prereg.md) |\n| `Q-xchan-at37-score` | ANSWERED | What does the @37 census say about the sealed X-channel offset pre-registration? | The census measured 1 - J = 0.020823, below every registered prediction, so scored exactly as registered every one of the seven candidates dies at \\|z\\| = 104 to 122 and the survivor set is EMPTY, an outcome the sealed prereg has no consequence clause for; the number itself survives an independent recount on a different marking scheme, and the offset question is replaced by a larger one. | X | [xchan-at37-score.md](history/staging/xchan-at37-score.md) |\n| `Q-xchannel-at23` | PARTIAL | What is the fifth point of the X-channel statistic, at @23, and does the monotone rise hold? | The fifth point is +0.2658 at @23, the rise holds at five levels and is larger than the four-point trend predicted, and the deficit has migrated into m >= 3, whose share of the X-gap runs 10.5, 23.5 and 81.1 per cent at @17, @19 and @23; the constant itself is not derived here. | X | [xchannel-at23.md](history/staging/xchannel-at23.md) |\n| `Q-xchannel-closedform` | PARTIAL | Does the zero-parameter closed form for the joint deficit survive its two blind tests at @29 and @31? | 1 - J = 4S2 is the only one of the three pre-registered laws left standing and it is not exact: a clean hit at @29, z = -0.90, and 4.93 sigma low at @31 where sigma_J falls to 0.000024, with the two rivals dying at 6.35 and 20.81 sigma; both blind levels agree on a stable relative offset of about half a percent, -0.41% and -0.48%, whose absolute residuals match to 2%. | X | [xchan-at29.md](history/staging/xchan-at29.md) |\n| `Q-xchannel-offset` | PARTIAL | What produces the ~3.8 law's -0.45% offset, and which candidate correction survives? | The best available description is a finite-level correction of S3's size and sign - 1 - J = 4S2 - S3 scores z = +0.04 at @29 and -0.41 at @31 where the standing law scores -0.43 and -2.32 - but it is POST HOC; two candidate corrections and one shape are refuted on existing data, the sigma behind the @31 detection is calibrated on seven known-truth controls, and the deciding blind test at @37 is pre-registered and committed alone. | X | [item-x-offset.md](history/staging/item-x-offset.md) |\n| `Q-xchannel-triples` | ANSWERED | Does the joint deficit deepen or decay as the mixed super-W triple census is extended past @17? | It decays: J = 0.9025, 0.9599, 0.9659 at @17, @19, @23, so the deficit runs 9.7%, 4.0%, 3.4%, both new points land inside the pre-registered DECAYING FASTER THAN FORCED band, and the DEEPENING instinct is refuted by 3.50 sigma between @17 and @19 and a further 1.68 sigma to @23. | X | [xchannel-triples.md](history/staging/xchannel-triples.md) |\n| `Q-y2-recompute` | ANSWERED | Does the sixteen-level Y2 ladder reproduce on the corrected instrument, and does Y2/x^2 fall? | The ladder reproduces exactly, sixteen of sixteen from sixteen independent processes, digit for digit against the published 2026-08-18 ladder, and Y2/x^2 falls at 15 of the 15 consecutive steps; this does NOT show that G2(x#)/x^2 falls, since the shortfall Y2/(G2-1) is unmeasured above x = 79. | 1d | [y2-recompute.md](history/staging/y2-recompute.md) |\n| `Q-z2-state` | PARTIAL | What is known about Z2, the largest gap between twin openers inside the zone (p, p'^2), and at what calibration? | Draft consolidation only, computes nothing new: the zone statement is an interval statement, the parity obstruction applies to it in full, and no proven statement about Z2 survives past the zone's own width. | none | [z2-state-draft-0829.md](history/staging/z2-state-draft-0829.md) |\n| `Q-z3-immune` | CLOSED | Do the square anchor's immune offset classes (stretch-01 §4) carry a survivor surplus beyond the redistribution-only guardrail, do they attract the floor's first survivor, and does restricting the transplanted caps to them lower K*? | No, on all three. The preregistered surplus test is VOID on its own matched control (a 0.6% ratio-of-means bias present with and without the quadratic point); the preregistered placement test HOLDS at Z = +0.49; the design-free class test reads the guardrail's r/(r-2) over 3.4 million twin openers with a largest per-prime deviation of 0.11% and per-prime Z in -0.48..+1.66; and the restricted certificate, while lowering K* at 61% of anchors, has no certificate at all at 93 of 1219 and is beaten head to head by the QR-BLIND restriction that certifies 1219 of 1219. The refinement is closed. | Z3 (retired) | [attack-z3-immune-01.md](history/staging/attack-z3-immune-01.md) |\n| `Q-zone-postulate` | OPEN | Is every zone occupied by a twin pair, and what is proven toward it? | OPEN: the weak form is PROVEN equivalent to the twin prime conjecture (both directions, elementary), the zone's supply is MEASURED to grow like p^2/ln^2 p with margin about p and verified by segmented sieve to 1e11, and the obstruction is named; none of the four attack routes closes it. | none | [ZONE-POSTULATE.md](ZONE-POSTULATE.md) |\n| `Q-zone-tail` | PARTIAL | What is the tail field of the zone (p, p'^2) - its law against ln^2, its worst case, its share of the R0 width, and is it the frozen-sieve last survivor below p'^2 at the rate the head is the frozen sqrt(p)-level first survivor above p? | The field exists now, 1,225 zones, and nothing in it is derived. The tail is the head's own renewal functional read at height p'^2 rather than p: c_tail = 0.7771 ln^2(p'^2) at [3163,1e4) against the head's 0.6693 ln^2 p on the same zones and the same estimator, both referenced to HL's 0.7574, and the apparent factor of four is the unit ln^2(p'^2) = 4 ln^2 p and nothing else. The band drift does not settle (0.7518, 0.7288, 0.7771; the pre-registered 0.03 step is missed), so this is band composition on two decades and not a measured convergence. The frozen-sieve mirror is 100% at the zone's own level p, but that is the Zone Restriction Lemma restated and not a rate comparable to the head's 96.65%; the non-vacuous mirror is one fold back, where the tail survives at 98.69% with the residual priced exactly (88 extra lpf(n) = p survivors, 18.18% of them biting), and the head's own frozen level sqrt(p) identifies the tail in only 22.86%, refuting this note's own pre-registered [0,5] zones. The renewal surplus t/R = 1.0619 at B4 is NOT RESOLVED, its bootstrap [0.9937, 1.1321] containing 1, and an exhaustive class-matched origin null over 5,962,057 origins gives the same 1.0619; three matched non-square endpoints see no difference at all, sign tests z = 0.14, -0.54, 0.66 with six paired intervals all containing zero, so nothing here is a property of the endpoint being a prime square. R0 is asserted at all 1,225 zones and the shares are 1.82% head, 89.72% Z2, 8.46% tail at B4 with the tail's share FALLING, so filling the tail row moves the decomposition's difficulty nowhere: Z2 was the obstruction before and is more of it after. | Z4 | [zone-tail-01.md](history/staging/zone-tail-01.md) |\n| `Q-zone-tail-02` | ANSWERED | On the full zonegap-01 range to 1e11, what is the per-zone tail law in both units, is the renewal surplus t/R real at 27,292 zones, and do the head and the tail of a zone have a joint law? | Resolved DOWN and nothing is derived: zone-tail-01's unresolved t/R = 1.0619 at 782 zones reads 1.0298 on a height-matched comparator at 17,700 zones (bootstrap [1.0156, 1.0443], excluding 1) and 1.0157 against the exactly matched discrete class null (bootstrap [1.0013, 1.0308], clearing 1 by 0.0013), so part of the original figure was the comparator and the pre-registered rule returns MATCH by two parts in ten thousand; c_local = 0.7522 at the top band with bootstrap [0.7410, 0.7630] now CONTAINING HL's 0.7574 and the band drift inside the registered 0.03 per step for the first time but still non-monotone (0.7518, 0.7664, 0.7416, 0.7522), so no convergence is measured; global worst tail/ln^2(p'^2) = 6.840 at p = 15107 with max/mean no longer flat (4.40, 5.36, 8.59, 8.43); Q5 is answered in the negative, head and tail showing no dependence at 27,292 zones (detrended r = -0.0009 against null s.e. 0.0061, permutation p = 0.8920, variance ratio 0.9990, no KS rejection, no chi-square near threshold), so the R0 pieces are a product structure to first and second moments and in the sum's whole law while independence is NOT proven and the simultaneous-extremes regime is untouched; a candidate rough-origin mechanism for the surplus is HEURISTIC and does not fit the head's own excess band by band; the head field's effective sample size is measured at 1,910 distinct a_first among 17,700 top-band zones, so this note's own head interval is priced over draws the field does not hold, a cluster bootstrap over distinct a_first (producer SEC I(a), seed 613009117) measuring the inflation at 4.10 rather than the 3.04 the redundancy alone implies and moving the top-band interval to [0.6848, 0.7692], still inside the registered [0.68, 0.78] (correction originating with redteam-0829-measure-b.md section 1c C15); the scope of that defect is that one interval and not the corpus, since destroyer-census-01.md publishes no bootstrap, head-residual-factor.js bootstraps over gaps, zonegap-01.js has none and zone-tail-01.js bootstraps only tail quantities; ten of ten predictions hit, which is a criticism of the pre-registration; and the R0 shares read 1.40% / 92.80% / 5.81% with the tail's share still falling, so Z2 is more of the obstruction after this pass than before it. | Z4 | [zone-tail-02-0829.md](history/staging/zone-tail-02-0829.md) |\n| `Q-zonegap-01-prereg` | SUPERSEDED | What is Z2(p) in the unswept half-decade p in [1e5, 316228)? | Pre-registration only, committed before the X = 1e11 sweep ran: five predictions are sealed on the per-zone distributional structure the published record ladders do NOT determine, with the records excluded from scoring and used as custody instead; the scoring document is the stage-2 embedded output, in zonegap-01.md. | none | [zonegap-01-prereg.md](history/staging/zonegap-01-prereg.md) |\n| `Q-zonegap-03-prereg` | OPEN | What is Z2(p) in the unswept decade X in (1e11, 1e12]? | Pre-registration only, sealed before any sweep past X = 1e11 exists: five theorem-grade predictions at sigma = 0 (resting on Z2(p) = env(p) identically, a two-line theorem conditional on the adopted ladder being the true running max) and five blind statistical ones with sigma models and kill rules, scored by the stage-3 embedded output with misses stated first. | none | [zonegap-03-prereg.md](history/staging/zonegap-03-prereg.md) |\n| `Q-zonegap-03-score` | ANSWERED | Do the ten predictions sealed in zonegap-03-prereg.md score against the stage-3 sweep at X = 1e12? | All ten sealed rows score HIT and none miss, after a second pass added a band-edge argument to zonegap-01.js and re-ran the decade so the two rows that named an unprinted band could be scored; the four Group T hits are a custody promotion that follows from CUSTODY 1 and 3 passing, one sub-clause of T5 (the full-decade sd) stays unprinted, and the sweep still needs a DERIVED engine because zonegap-01.js's inlined 41-record ladder makes it exit at 1e12. | Z7 | [zonegap-03-score.md](history/staging/zonegap-03-score.md) |\n| `Q-zonegap-Z2` | PARTIAL | What is Z2(p), the largest twin-pair gap inside the zone (p, p′^2), measured at every zone to 1e11? | MEASURED to 1e11 over 27,292 zones: every zone holds at least two twin pairs, Z2 is the record envelope wearing zone coordinates, Z2(p) <= G2(p#) is a one-line theorem with equality only to p = 7, and the zone-local margin runs away like p^2/(3.9 ln^3 p); the pre-registration scored 2 of 5. | none | [zonegap-01.md](history/staging/zonegap-01.md) |\n| `Q-zonegap-model` | ANSWERED | What is Z2 once decomposed, and does the per-zone field carry anything the record ladder does not? | Z2(p) = env(p) with D(p) identically zero (PROVEN conditional on the adopted ladder, VERIFIED 204 ways), so the per-zone field is the record ladder wearing zone coordinates and the u bimodality is closed; the blind tests for the [1e11, 1e12] decade are sealed and the sweep itself is not run. | none | [zonegap-03-model.md](history/staging/zonegap-03-model.md) |\n| `Q-zonegap-prior-art` | ANSWERED | Who owns Z2(p), the maximal twin-prime gap below a quadratic cutoff, in the owning conventions? | Kourbatov (2013) and Kourbatov-Wolf (2019) own maximal twin-prime gaps, with heuristics and records and nothing proven; the zone form (p, p′^2) has five neighbours and no owner, and the per-zone quadratic-cutoff object appears absent on channels calibrated in the same session. | none | [zonegap-prior-art.md](history/staging/zonegap-prior-art.md) |\n| `Q-zonegap-reduction` | PARTIAL | What must be proven for Z2, what already is, and what does the proven head machinery give the zone? | The Zone Restriction Lemma is proven and verified at five levels - the in-zone twin slots of T_p ARE the in-zone twin primes, so Z2(p) IS the whole-tile gap object restricted to the head window - the three-slack chain is graded honestly, the transfer inventory is listed item by item, and the deep end is proven an onset desert over 24 levels; section 5 lists what was not reached. | Z (retired) | [zonegap-02-reduction.md](history/staging/zonegap-02-reduction.md) |\n| `Q-zonegap-witnesses` | ANSWERED | Do independent second witnesses confirm the twin-gap record table, and does the due OEIS flip-search turn anything up? | Both witnesses AGREE at 82/82 rows with no extension and no conflict and the lesser-to-lesser convention verified rather than assumed; the OEIS check returns one loud hit, the per-zone pair-count sequence being A273257 one index over, while the gap objects stay absent, and one literature item, Korevaar 2012, is still walled and read only by arXiv proxy. | none | [zonegap-witnesses.md](history/staging/zonegap-witnesses.md) |\n\n## 3. Unindexed notes (0): title only, until a block is added\n\n*(Legacy notes written before the ledger existed. A grep for the object still\nlands here. Add a block when a note is next touched; `node research/qc.js\nledger-backlog` lists them.)*\n\n", "research/qc/questions.js": "// ============================================================================\n// QC / QUESTIONS — which questions has this corpus already attacked, and where?\n// ============================================================================\n// Chris, 2026-08-28: \"We keep re-running things. Restructure repo so that does\n// not happen.\" The same day three briefs re-posed questions the corpus had\n// already answered: TODO item 0's first move had run a week earlier\n// (attack-rhoms-01.md), item 1c's refit had run 2026-08-21, and a TODO premise\n// in item 10 had been measured under another name. 316 staging notes answered\n// questions and nothing indexed them BY QUESTION; the router indexes by\n// document and the changelog by claim, and a briefer greps on guessed words.\n//\n// THE MECHANISM. Every note that answers a question carries a machine-readable\n// block at its top, invisible when rendered:\n//\n// \n//\n// THE PARITY GATE (2026-09-05), scoped after the independent review.\n// A note declares its arithmetic inputs and method scope, or `residue-only`.\n// A mathematical obstruction must separately name the retained statistics,\n// error tolerance and quantifiers; the gate does not prove impossibility.\n// Exact residue arrangements encode primality. Merely using residue notation\n// does not put every argument in one sieve-method class.\n// The check rejects a missing or placeholder line on a new attack note whose\n// todo names a live item. Legacy notes are exempt by git's add date; an\n// untracked note is new. `residue-only` remains a supported declaration.\n//\n\n// Several notes may share an id. Their statuses are reported as one status only\n// when they agree; otherwise the index prints MIXED with each note's status.\n// The verdict shown is the last note's in path order; the full chain is listed.\n//\n// research/QUESTIONS.md is GENERATED from these blocks (node research/qc.js\n// --index, or node research/gen-questions-index.js) and is the one place to\n// look before briefing: by TODO item, then by question id. Notes without a\n// block are listed there by title under \"unindexed\", so a grep finds them too.\n//\n// THE GUARD, gated: a TODO item must list, on a line `Ledger: Q-a, Q-b`,\n// every question id whose block names that item. A move cannot be re-briefed\n// without reading, in the item itself, that it has run. The gate says so when\n// a note names an item that does not acknowledge it.\n//\n// The status belongs to the QUESTION, not the note's review grade. An answered\n// specification may still contain an open arithmetic target. Read the owning\n// note and linked reviews for its grade; the generated preamble explains this.\n// ============================================================================\n'use strict';\nconst fs = require('fs');\nconst path = require('path');\nconst C = require('./corpus');\n\nconst STATUSES = new Set(['OPEN', 'PARTIAL', 'ANSWERED', 'CLOSED', 'SUPERSEDED']);\nconst BLOCK_RE = //;\nconst TODO_PATH = path.join(C.ROOT, 'TODO.md');\nconst OUT_PATH = path.join(C.RESEARCH, 'QUESTIONS.md');\n\nfunction indexedFiles() {\n // Staging notes and the working documents under research/. Registries carry\n // a `Q-registry-` block (their question is what they index); generated\n // files (SCRIPTS.md, QUESTIONS.md) carry none and are excluded by corpus.js. History other than staging is never\n // audited (qc/README.md), and that boundary holds here too.\n const staging = C.historyMarkdown.filter(f => /history\\/staging\\//.test(f));\n const body = C.bodyMarkdown.filter(f => /^research\\/[^/]+\\.md$/.test(C.rel(f)));\n return [...staging, ...body].sort();\n}\n\nfunction parseBlock(file) {\n const src = C.read(file);\n const head = src.split('\\n').slice(0, 60).join('\\n');\n const m = BLOCK_RE.exec(head);\n const title = (src.split('\\n').find(l => /^#\\s/.test(l)) || '').replace(/^#\\s*/, '').trim();\n if (!m) return { file, title, block: null };\n const fields = {};\n const problems = [];\n for (const raw of m[1].split('\\n')) {\n const ln = raw.trim();\n if (!ln) continue;\n const km = /^([a-z]+):\\s*(.*)$/.exec(ln);\n if (!km) { problems.push(`unparseable line \"${ln}\"`); continue; }\n fields[km[1]] = km[2].trim();\n }\n for (const k of ['id', 'status', 'question', 'verdict'])\n if (!fields[k]) problems.push(`missing field \"${k}\"`);\n if (fields.status && !STATUSES.has(fields.status))\n problems.push(`status \"${fields.status}\" is not one of ${[...STATUSES].join(' | ')}`);\n if (fields.id && !/^Q-[A-Za-z0-9][A-Za-z0-9-]*$/.test(fields.id))\n problems.push(`id \"${fields.id}\" must look like Q-slug (letters, digits, hyphens)`);\n const todo = (fields.todo || 'none').split(',').map(s => s.trim()).filter(s => s && s !== 'none')\n .map(s => ({ id: s.replace(/\\s*\\(retired\\)\\s*$/, ''), retired: /\\(retired\\)/.test(s) }));\n const line = src.slice(0, m.index).split('\\n').length;\n return { file, title, block: { ...fields, todo, line }, problems };\n}\n\nfunction todoItems() {\n // A corpus without a TODO.md has no items to acknowledge; the block side of the\n // check still runs. The selftest fixture carries its own TODO.md.\n if (!fs.existsSync(TODO_PATH)) return new Map();\n const src = fs.readFileSync(TODO_PATH, 'utf8').split('\\n');\n const items = new Map();\n let cur = null;\n src.forEach((ln, i) => {\n const h = /^\\*\\*([A-Za-z0-9]+)\\.\\s/.exec(ln);\n if (h) { cur = { id: h[1], line: i + 1, ledger: new Set(), ledgerLine: null }; items.set(h[1], cur); return; }\n if (/^##\\s/.test(ln)) { cur = null; return; }\n if (cur) {\n const lm = /^\\s*Ledger:\\s*(.*)$/.exec(ln);\n if (lm) { cur.ledgerLine = i + 1; for (const q of lm[1].matchAll(/Q-[A-Za-z0-9-]+/g)) cur.ledger.add(q[0]); }\n }\n });\n return items;\n}\n\nfunction collect() {\n const rows = indexedFiles().map(parseBlock);\n return { rows, todo: todoItems() };\n}\n\n// ---------------------------------------------------------------------------\n// THE CHECK\n// ---------------------------------------------------------------------------\nfunction ledger() {\n const { rows, todo } = collect();\n const findings = [];\n const byId = new Map();\n for (const r of rows) {\n if (!r.block) continue;\n for (const p of r.problems || [])\n findings.push({ file: C.rel(r.file), line: r.block.line, kind: 'ledger-malformed', detail: p,\n note: 'the block format is documented at the top of qc/questions.js' });\n if (r.block.id) {\n if (!byId.has(r.block.id)) byId.set(r.block.id, []);\n byId.get(r.block.id).push(r);\n }\n for (const t of r.block.todo) {\n if (t.retired) continue;\n const item = todo.get(t.id);\n if (!item) {\n findings.push({ file: C.rel(r.file), line: r.block.line, kind: 'ledger-todo-unknown',\n detail: `todo: ${t.id} names no item heading in TODO.md`,\n note: 'if the item has left TODO.md, write \"todo: ' + t.id + ' (retired)\"; otherwise fix the id' });\n continue;\n }\n if (!item.ledger.has(r.block.id)) {\n findings.push({ file: 'TODO.md', line: item.line, kind: 'ledger-todo-unlisted',\n detail: `item ${t.id} does not list ${r.block.id} (${C.rel(r.file)}, ${r.block.status}) on its Ledger: line`,\n note: 'THE RE-RUN GUARD: the item must acknowledge every question already attacked under it. Add the id to the item\\'s \"Ledger:\" line, and read the note before briefing any move' });\n }\n }\n }\n for (const [id, group] of byId) {\n const qs = new Set(group.map(g => C.norm(g.block.question)));\n if (qs.size > 1)\n findings.push({ file: C.rel(group[1].file), line: group[1].block.line, kind: 'ledger-id-conflict',\n detail: `id ${id} is shared by ${group.length} notes with different question text`,\n note: 'notes sharing an id answer the SAME question; give a different question a different id' });\n }\n for (const [tid, item] of todo)\n for (const q of item.ledger)\n if (!byId.has(q))\n findings.push({ file: 'TODO.md', line: item.ledgerLine || item.line, kind: 'ledger-todo-dangling',\n detail: `item ${tid} lists ${q}, which no note carries`,\n note: 'a Ledger: line points at notes that exist; fix the id or add the block to the note' });\n const indexed = rows.filter(r => r.block).length;\n return {\n name: 'ledger',\n description: `every note that answers a question says which, and every TODO item acknowledges the questions already attacked under it `\n + `(${indexed} of ${rows.length} notes carry a ledger block; ${byId.size} question ids; ${todo.size} TODO items)`,\n findings,\n };\n}\n\n// ---------------------------------------------------------------------------\n// THE PARITY GATE (2026-09-05). See the header.\n// ---------------------------------------------------------------------------\nconst PARITY_CUTOFF = '2026-09-05';\nconst PARITY_EXEMPT = 'residue-only';\nlet addDates = null;\nfunction fileAddDates() {\n // First-commit date of every file under research/, one git call, cached.\n // A corpus without git (the selftest fixture) yields an empty map, and an\n // unknown file counts as NEW, which is the conservative direction: the gate\n // asks a question rather than assuming the answer.\n if (addDates) return addDates;\n addDates = new Map();\n try {\n const out = require('child_process').execFileSync('git',\n ['-C', C.ROOT, 'log', '--diff-filter=A', '--format=%as', '--name-only', '--', 'research'],\n { encoding: 'utf8', stdio: ['ignore', 'pipe', 'ignore'] });\n let date = null;\n for (const ln of out.split('\\n')) {\n if (/^\\d{4}-\\d\\d-\\d\\d$/.test(ln)) { date = ln; continue; }\n const f = ln.trim();\n if (!f || !date) continue;\n // log is newest first; the earliest add wins (a file re-added after a\n // rename keeps its original date)\n addDates.set(f, date);\n }\n } catch (e) { /* no git: every file is new */ }\n return addDates;\n}\n\nfunction parity() {\n const { rows } = collect();\n const dates = fileAddDates();\n const findings = [];\n let attacks = 0, exempt = 0;\n for (const r of rows) {\n if (!r.block) continue;\n const live = r.block.todo.filter(t => !t.retired);\n if (!live.length) continue; // not an attack note\n const rel = C.rel(r.file);\n const added = dates.get(rel);\n if (added && added < PARITY_CUTOFF) continue; // legacy note, exempt\n attacks++;\n const v = (r.block.parity || '').trim();\n if (!v) {\n findings.push({ file: rel, line: r.block.line, kind: 'parity-missing',\n detail: `attack note (todo: ${live.map(t => t.id).join(', ')}) carries no parity: line`,\n note: 'THE PARITY GATE: state the arithmetic inputs and method scope, or write \"parity: residue-only\". This checks documentation, not mathematical feasibility. Format at the top of qc/questions.js' });\n continue;\n }\n if (v.toLowerCase() === PARITY_EXEMPT) { exempt++; continue; }\n if (v.length < 12 || /^(n\\/?a|none|tbd|todo|unknown|\\?+|-+)$/i.test(v)) {\n findings.push({ file: rel, line: r.block.line, kind: 'parity-placeholder',\n detail: `parity: \"${v}\" names no input`,\n note: 'either state the arithmetic inputs and method scope in one line, or write exactly \"residue-only\"' });\n }\n }\n return {\n name: 'parity',\n description: `every attack note added from ${PARITY_CUTOFF} states its arithmetic inputs and method scope, or declares residue-only `\n + `(${attacks} attack notes in scope, ${exempt} declared residue-only)`,\n findings,\n };\n}\n\nfunction ledgerBacklog() {\n const { rows } = collect();\n const missing = rows.filter(r => !r.block);\n const findings = missing.map(r => ({ file: C.rel(r.file), line: 1, kind: 'ledger-missing',\n detail: r.title ? `\"${r.title.slice(0, 90)}\"` : '(no title)',\n note: 'legacy note without a ledger block; it is listed by title in QUESTIONS.md §unindexed until one is added' }));\n return {\n name: 'ledger-backlog',\n description: `notes with no ledger block, listed by title in research/QUESTIONS.md so a grep still finds them (${missing.length} of ${rows.length}; the counter should only fall)`,\n findings,\n };\n}\n\n// ---------------------------------------------------------------------------\n// THE GENERATOR — research/QUESTIONS.md\n// ---------------------------------------------------------------------------\nconst cell = s => String(s || '').replace(/\\|/g, '\\\\|').replace(/\\n/g, ' ').trim();\nconst link = f => { const r = C.rel(f); const p = r.replace(/^research\\//, ''); return `[${path.basename(r)}](${p})`; };\n\nfunction generate() {\n const { rows, todo } = collect();\n const withBlock = rows.filter(r => r.block && r.block.id);\n const byId = new Map();\n for (const r of withBlock) { if (!byId.has(r.block.id)) byId.set(r.block.id, []); byId.get(r.block.id).push(r); }\n const ids = [...byId.keys()].sort();\n const summarise = id => {\n const g = byId.get(id);\n // Deterministic display choice: the last note in path order supplies the\n // verdict. This is NOT chronology or review precedence. List every record\n // and explain the choice in the generated preamble.\n const last = g[g.length - 1].block;\n // A shared id carries one status ONLY when its notes agree. When they do not,\n // the index says MIXED and lists each note's own status, so a reader opens\n // them instead of trusting a rank. (2026-08-28 red team: taking the highest\n // rank showed Q-c2prime-drift as CLOSED while its earlier note reads PARTIAL\n // and the later one closed on a weaker window.)\n const statuses = [...new Set(g.map(x => x.block.status))];\n const status = statuses.length === 1 ? statuses[0]\n : 'MIXED (' + g.map(x => path.basename(x.file) + ': ' + x.block.status).join('; ') + ')';\n return { id, question: last.question, status, verdict: last.verdict, files: g.map(x => x.file), todo: [...new Set(g.flatMap(x => x.block.todo.map(t => t.id + (t.retired ? ' (retired)' : ''))))] };\n };\n const S = ids.map(summarise);\n\n const out = [];\n out.push('# Questions: what has been asked here, what it got, and where the record is');\n out.push('');\n out.push('*(GENERATED from the `` blocks at the top of the notes; never');\n out.push('hand-edited. Regenerate with `node research/qc.js --index` or');\n out.push('`node research/gen-questions-index.js`. The block format and the gate that');\n out.push('enforces it are documented at the top of `qc/questions.js`.)*');\n out.push('');\n out.push('**Before proposing an attack or briefing an agent:** find the TODO item in');\n out.push('section 1, search for the actual object, then read the relevant records and');\n out.push('their decisive dependencies. Use [OUTCOMES.md](OUTCOMES.md) for reuse and');\n out.push('revisit conditions. An item can span many unrelated attempts; loading every');\n out.push('linked note is not the required first step. Verify any conclusion that');\n out.push('determines the next move, including an ANSWERED or CLOSED entry.');\n out.push('');\n out.push('**Status describes the question, not proof strength or review grade.**');\n out.push('OPEN: unresolved. PARTIAL: specified parts are resolved. ANSWERED: the bounded');\n out.push('question has an answer, which may be a negative finding, finite measurement');\n out.push('or completed specification. Its underlying arithmetic target may remain OPEN.');\n out.push('CLOSED: the stated premise or approach is closed at its recorded scope; this');\n out.push('does not refute every related conjecture. SUPERSEDED: another record replaces');\n out.push('this one. Consult the owning note and linked reviews for HELD or other grades.');\n out.push('');\n out.push('**Multiple records:** conflicting statuses are shown as MIXED. The displayed');\n out.push('verdict comes from the last record in path order, not necessarily the newest');\n out.push('or most authoritative review. Read the linked records before resolving a');\n out.push('disagreement; the generator does not adjudicate it.');\n out.push('');\n out.push('## 1. By TODO item: what has already run under each');\n out.push('');\n out.push('| TODO item | question | status | verdict | records |');\n out.push('|---|---|---|---|---|');\n const itemOrder = [...todo.keys()];\n const seen = new Set();\n for (const tid of itemOrder) {\n const here = S.filter(s => s.todo.some(t => t.replace(/ \\(retired\\)$/, '') === tid));\n for (const s of here) { seen.add(s.id); out.push(`| ${tid} | \\`${s.id}\\` ${cell(s.question)} | ${s.status} | ${cell(s.verdict)} | ${s.files.map(link).join(', ')} |`); }\n }\n const retired = S.filter(s => !seen.has(s.id) && s.todo.length);\n for (const s of retired) out.push(`| ${cell(s.todo.join(', '))} | \\`${s.id}\\` ${cell(s.question)} | ${s.status} | ${cell(s.verdict)} | ${s.files.map(link).join(', ')} |`);\n out.push('');\n out.push('## 2. Every question, by id');\n out.push('');\n out.push('| id | status | question | verdict | TODO | records |');\n out.push('|---|---|---|---|---|---|');\n for (const s of S) out.push(`| \\`${s.id}\\` | ${s.status} | ${cell(s.question)} | ${cell(s.verdict)} | ${cell(s.todo.join(', ') || 'none')} | ${s.files.map(link).join(', ')} |`);\n out.push('');\n const missing = rows.filter(r => !r.block);\n out.push(`## 3. Unindexed notes (${missing.length}): title only, until a block is added`);\n out.push('');\n out.push('*(Legacy notes written before the ledger existed. A grep for the object still');\n out.push('lands here. Add a block when a note is next touched; `node research/qc.js');\n out.push('ledger-backlog` lists them.)*');\n out.push('');\n for (const r of missing) out.push(`- ${link(r.file)}: ${cell(r.title)}`);\n out.push('');\n fs.writeFileSync(OUT_PATH, out.join('\\n'));\n return { questions: S.length, notes: withBlock.length, unindexed: missing.length };\n}\n\nmodule.exports = { parity, ledger, ledgerBacklog, generate, collect };\n\nif (require.main === module) {\n const r = generate();\n console.log(`research/QUESTIONS.md: ${r.questions} questions from ${r.notes} indexed notes, ${r.unindexed} unindexed`);\n}\n", "research/history/staging/xchan-at37-score.md": "# The @37 prereg scored: every registered candidate is killed, the number survives an independent recount, and the offset question is replaced by a larger one\n\n\n\n*(2026-08-21. HELD for one adversarial pass before anything downstream quotes\nit. Scores the SEALED pre-registration\n[xchan-at37-offset-prereg.md](xchan-at37-offset-prereg.md) (committed alone at\n`0a7dd73`) against the @37 census `research/xchan-at37-01-census.js` (embedded\n2026-08-21, 21354.0 s, out-sha256 `6d83c928…`), then verifies the surprising\nnumber with an independent implementation,\n`research/xchan-at37-02-verify.js` (formally embedded, 335.1 s). Every figure\nbelow is from an embedded OUTPUT block or one-line arithmetic on figures in\none, inputs shown. Grades are explicit throughout: MEASURED, REGISTERED\nVERDICT, INTERPRETATION.)*\n\n## Verdict\n\n**The census measured 1 − J = 0.020823 at @37 — below every prediction the\nprereg registered. Scored exactly as registered (z on the run's own\nslot-clustered σ, |z| > 3 kills), every one of the seven candidates dies, at\n|z| = 104–122; the attack-sigma31 calibration bracket [0.745, 2.293] moves\nthe weakest reading to 45σ and changes no verdict. The survivor set is EMPTY\n— an outcome the sealed prereg has no consequence clause for. And the number\nis right: an independent implementation (different marking scheme,\nprime-value divisor lists, BigInt-exact U and classification samples), gated\nbyte-exact on the full @23 period against the natal-cap-39 reference, agrees\nwith the producer's own kernel field-for-field — all 17 compared fields EQUAL\n— on the contiguous prefix [0, 3e10) of the @37 period. The offset the\nprereg was built to adjudicate (a ~1.2e−4 flat absolute residual) is gone as\na question: the residual at @37 is 1.04e−3 — nine times larger, 13.4·S₃,\nbelow the entire finite-level-correction family including both members\nalready refuted at @31 for subtracting too much.** What stands, what died,\nand what it means:\n\n1. **[REGISTERED VERDICT] All seven candidates KILLED; no clause fires.** The\n prereg's three consequence clauses each presuppose at least one survivor;\n none is met. The J-band clause (J outside (0.94, 1.00] ⇒ \"the object\n changed\") does NOT fire: J = 0.979177. P1 (aligned super-W = 0): HIT. The\n validation gate: @23 ALL PASS, @31 reproduced digit-exact — the value is\n reportable under the prereg's own §3.\n2. **[MEASURED] The instrument's error-bar model held; the world moved.** The\n run's own σ_slot = 8.511e−6 sits inside the prereg's projection band\n [8.44e−6, 8.64e−6], and the exact tail mass (0.070749) equals the\n registered arithmetic. Nothing about the σ machinery failed — the\n measurement fell 122 of those σ outside the nearest registered candidate.\n3. **[MEASURED] The recount agrees exactly.** Slot count, B3 four-way split,\n sub/sup per class, per-slot subtotals ΣX/ΣX², maxima: all EQUAL between\n the independent walker and the producer's unmodified kernel on the same\n range. The 2^53 close calls are audited exact in BigInt (C(K,3) at 86.5%\n of 2^53; the producer never forms q₁q₂q₃ — its U-test compares\n q₁q₂ ≤ y² = 7.42e12 = 0.08% of 2^53).\n4. **[INTERPRETATION] At least one of the two structural trends bends hard at\n @37.** Under item-x-offset's surplus/deficit split, the measured net\n forces either the sieve-deficit ratio D/4S₂ to fall 1.18 → 1.15 → ~1.09,\n or the line bias β to stop decaying (β/S₃ jumping ~41 → 48–57). No @37\n line census exists, so the split is one equation in two unknowns — the\n line ensemble at @37 is now the decisive missing instrument.\n5. **[INTERPRETATION] The \"asymptote exactly 4\" reading dies in its monotone\n form; the ~3.8 law survives only as a ≈5% scale law.** Measured (1−J)/F:\n 3.8242 (@29), 3.8427 (@31), 3.6907 (@37) — the approach to 4 reverses,\n with a one-level drop 4× the size of the prior two-level climb.\n\n## 1. The registered score\n\nThe statistic, the σ, and the rule are the prereg's §3, verbatim: z_c =\n((1−J)_meas − pred_c)/σ_slot on the run's own slot-clustered σ; |z_c| > 3\nKILLED. Measured (census OUTPUT block): obs = 60,388,809,837,\nCRT = 61,673,023,269.22, 1 − J = 0.020823 (exact 0.020822936),\nσ_reg = 3.9846e−6, σ_slot = σ_reg × 2.136 = 8.5111e−6 (prints as the block's\nrounded 0.000009; inside the prereg's projection band). Scoring table, from\nthe embedded OUTPUT of `research/xchan-at37-02-verify.js`:\n\n| candidate | registered 1−J(37) | z (σ_slot) | registered verdict | at the @31 calibration bracket [0.745, 2.293] |\n|---|---|---|---|---|\n| N1 `4S₂` | 0.021863 | **−122.2** | KILLED | reads −53.3 to −164.0 — verdict unchanged |\n| C1 `4S₂ − S₃` | 0.021785 | −113.0 | KILLED | −49.3 to −151.7 — unchanged |\n| C2 `4S₂ − 2S₃` | 0.021707 | −103.9 | KILLED | −45.3 to −139.4 — unchanged |\n| M-mult | 0.021759 | −110.0 | KILLED | −48.0 to −147.6 — unchanged |\n| M-abs | 0.021744 | −108.2 | KILLED | −47.2 to −145.3 — unchanged |\n| M-cS3 | 0.021770 | −111.3 | KILLED | −48.5 to −149.4 — unchanged |\n| M-ln | 0.021772 | −111.5 | KILLED | −48.6 to −149.7 — unchanged |\n\n**FAMILY VERDICT: no survivors.** For context only (already REFUTED at @31\nand not carried by the prereg): C4 scores −85.7 and C8 −51.6 — the measured\npoint sits below the entire priced family, including the members killed at\n@31 for subtracting too much.\n\nRegistered secondaries: **P1 HIT** (aligned super-W obs = 0). **The J-band\nclause does not fire** (J = 0.979177 ∈ (0.94, 1.00]). **Gate PASS** (@23\nreference ALL PASS — 14 integers + 3 rounded reals; the @31 blind row\nreproduced to the digit: obs 1653241687, 1−J 0.024666, z(slot) −2.32,\nd −0.48%). **N̄ counted = formula** (145,286,237,250). **Missing mass =\n0.070749**, equal to the registered exact value\n(`research/attack-x-offset-01-terms.js`). **The line-bias secondary was not\nrun** — the census is the natal producer only, and the prereg registered the\nline ensemble as deferrable (3–5× the natal cost); there is no β(37) to\nread, and §3 below feels its absence. Orientation symmetry holds:\n(2,1) − (1,2) = −41185 of 6.04e10, −0.2·√obs.\n\n### What the prereg prescribes for this outcome, checked against its wording\n\nPrereg §3 defines KILLED/ALIVE per candidate and says \"the family verdict\nnames every survivor, sorted by |z_c|\" — an empty survivor list is\nrepresentable, and this file records exactly that. But its three consequence\nclauses are each conditioned on a survivor pattern (\"if N1 survives…\", \"if\nN1 is killed and at least one shrinking-family member survives…\", \"if only\nM-abs survives…\"): none is met. The one registered catch-all for a wild\nvalue — the J-band clause — does not fire, because J stayed inside\n(0.94, 1.00]. **The sealed prereg therefore prescribes nothing for this\noutcome beyond what is recorded above.** TODO item X's offset clause cannot\nbe updated under any registered branch; a fresh proposal is in §5, and it is\na proposal only.\n\n## 2. The independent recount — the number is right\n\n*Producer: `research/xchan-at37-02-verify.js`, formally embedded (335.1 s).\nThe discrepancy to the standing law N1 is 0.106% of obs\n(6.4e7 of 6.04e10 triples; 0.090% to the nearest candidate, C2), so a\nrange-restricted recount resolves it; the\nrange was priced by pilot before the run (predicted ~208 s + ~90 s, measured\n251.4 s + 81.9 s).*\n\n**Independence, by construction.** The verifier shares no kernel with the\nproducer: marking is by direct multiple-enumeration in v-space (walk every\nmultiple of every prime, classify by m mod 30 tracked incrementally — no CRT\nresidues, no modular inverse, no compressed-index stride); the per-slot\ndivisor lists store the PRIME VALUE in Uint32, not an index into a prime\narray, so the index-alias class of the repaired 2026-08-21 census defect\ncannot exist in it; U[q] = ⌊W/q⌋ comes from BigInt division; a deterministic\n1-in-2^20 sample of triple classifications is re-decided in BigInt as\nq₁q₂q₃ vs W. Its prime list and level constants are rebuilt from scratch and\nchecked (K = 198274, y = 2724079, W and N̄ recomputed in BigInt).\n\n**Gates.** [MEASURED] On the full @23 period both the independent engine and\nthe producer's kernel — loaded unmodified from its file at run time and\nrange-restricted through W, which `census()` reads only as kmax = W/30 —\nreproduce the natal-cap-39 reference on all 11 compared figures: ALL PASS,\ntwice.\n\n**The recount.** [MEASURED] On the contiguous prefix [0, 3e10) — 0.4043% of\nthe @37 period — the two implementations agree EXACTLY on all 17 compared\nfields: natal slot count 587,352,241; the B3 four-way split; all four sub-W\nclasses; all four super-W classes (both aligned classes 0 — P1 holds on the\nprefix); the per-slot subtotals ΣX = 244,865,942 (the prefix's mixed\nsuper-W obs) and ΣX²; and both divisor-list maxima. 1753 sampled BigInt\nclassification re-decisions: 0 disagree. The U-table audit: all 198,274\nproducer entries equal the BigInt floor. The prefix's obs density sits 0.30%\nabove the full-period average (informative; density varies along the line).\n\n**The 2^53 audit the width lesson demanded.** [MEASURED] C(K,3) =\n1,299,090,727,729,024 in BigInt equals the printed figure, and the float\npath is exact — K(K−1)(K−2) = 7.79e15 is 86.5% of 2^53 and still below it.\nTriple-product formation: the producer never multiplies three primes; its\nU-test compares q₁q₂ against ⌊W/q₃⌋, and max q₁q₂ ≤ y² = 7.42e12 = 0.08% of\n2^53. The verifier's own accumulators peak at 8.9e−6% of 2^53.\n\n**RECOUNT VERDICT: byte-exact agreement on the prefix — the counting is\nright and 1 − J = 0.020823 stands.** What the prefix does not certify: a\ndefect expressed only beyond 3e10 of the walk. No position-scaled container\nin the producer crosses 2^53 before @43 (the width-sweep table plus the\naudit above), so the residual exposure is the one every full-period count\ncarries.\n\n## 3. Interpretation, through the surplus/deficit split\n\n*Everything here is INTERPRETATION over MEASURED anchors; grades marked.\nThe anchors (census OUTPUT; `attack-x-offset-01-terms.js` OUTPUT; ratios are\none-line arithmetic on them, inputs shown):*\n\n| x | 4S₂ | measured 1−J | Δ = 4S₂−(1−J) | d | Δ/S₃ | (1−J)/F |\n|---|---|---|---|---|---|---|\n| 23 | 0.033678 | 0.034068 | −3.90e−4 | +1.16% | — | 3.8630 (noisy level) |\n| 29 | 0.028943 | 0.028823 | +1.20e−4 | −0.41% | 0.92 | 3.8242 |\n| 31 | 0.024784 | 0.024666 | +1.19e−4 | −0.48% | 1.21 | 3.8427 |\n| 37 | 0.021863 | 0.020823 | **+1.04e−3** | **−4.76%** | **13.4** | **3.6907** |\n\n**[MEASURED] The offset stopped being an offset.** Two sharp levels carried\nthe same absolute residual (~1.19e−4); @37's is 8.8× larger. A correction\nfamily built to shrink (S₃-scale) or stay flat (constant absolute) cannot\nfollow a residual that grows nine-fold in one level while its own scale\nfalls.\n\n**[INTERPRETATION] Through item-x-offset's decomposition.** The natal J is\nthe net of a positive, decaying lattice-line surplus (β = J_line − 1 = 9.88,\n7.44, 5.38, 4.03 e−3 over @19..@31; β/S₃ = 39–43, a sighting) and a deeper\nsieve deficit (D = 1 − J/J_line = 1.18× and 1.15× 4S₂ at @29/@31). The @37\nrun measured only the net: one equation, two unknowns.\n\n- If β continues the β/S₃ ≈ 39–43 sighting (S₃(37) = 7.78e−5), then\n β(37) ≈ 3.04–3.35e−3 and the implied deficit is D = 0.0238–0.0241, i.e.\n D/4S₂ ≈ 1.09–1.10: the sieve deficit's excess over 4S₂ would be collapsing\n (1.18 → 1.15 → ~1.09).\n- If instead D/4S₂ eases along its prior trend to 1.12–1.15, the implied\n β(37) is 3.76–4.43e−3, i.e. β/S₃ = 48–57: the line bias's decay stalls or\n reverses.\n\nEither the deficit ratio or the β decay breaks trend; the measured net\ncannot say which. **The @37 line ensemble is now the decisive missing\nmeasurement** (registered as deferrable in the prereg at 3–5× the natal\ncost, ~17–29 h on the existing instrument).\n\n**[INTERPRETATION] A mechanism with the right sign, growth and provenance is\nalready on the record.** item-x-offset §6b DERIVED the cofactor-one surplus:\nnatal enrichment of the far-super-W layers by ∏_{p≤x} p/(p−1) → e^γ ln x — a\nsurplus component that GROWS with the level, pushes J up against the bulk\ndeficit, and is verified to 0.001% at @31. A growing surplus is exactly what\na nine-fold one-level shortfall of the net deficit needs. Whether its\naggregate CRT-mass weight at @37 is the right size is a closed-form\ncomputation over prime sums (extend `attack-x-offset-03-cofactor.js` to\n@37), not a census — the cheapest sharp test on the table.\n\n**[INTERPRETATION] What survives, dies, transforms.**\n\n- The \"asymptote exactly 4\" reading DIES in its monotone form: the measured\n (1−J)/F ladder turns down (3.8242 → 3.8427 → 3.6907), and the one-level\n drop is 4× the prior climb. A non-monotone approach to 4 is not excluded —\n nothing constrains it — but the reading as held is gone.\n- The ~3.8 law TRANSFORMS: dead as a sharp law (−4.76% at 122 σ_slot), alive\n as a scale law (1−J within 5% of 4S₂ across @29–@37). 4S₂ now reads as an\n upper envelope with an accelerating downward departure.\n- The 1.15–1.18 multiplicative split is UNDETERMINED at @37, not refuted:\n both of its factors cannot keep their trends simultaneously (above), and\n deciding which one bent requires β(37).\n- The offset-correction family (the prereg's whole candidate space) DIES as\n a family: the needed correction at @37 is 13.4·S₃ — bigger than C4 and C8,\n the two members already refuted at @31 for being too big. No finite-level\n correction with a shrinking or constant scale fits @29, @31 and @37\n simultaneously.\n\n**[INTERPRETATION] What the σ-calibration finding means for the 122σ.** The\nregistered z's are verdicts on the registered predictions, and no\ncalibration inside the measured bracket rescues any candidate. But 122σ is\nnot 122σ of sampling surprise: attack-sigma31 measured the raw\n(non-cancelling) statistic's level-scale arithmetic term at 82 σ_slot at @31\n— common to the whole window, invisible to any within-window control\nensemble, uncontrolled at every new level because each level owns exactly\none window. The @37 deviation, 1.04e−3 in J-units, is about a quarter of\n@31's own line bias (4.03e−3): comfortably inside the scale that level\narithmetic is KNOWN to produce on the raw form. The honest reading of any\nraw-form z at a new level is that the sampling σ is not the yardstick for\ncross-level model error — a future prereg that wants probabilistic teeth at\na new level must register an arithmetic-term (β) model, which the @37\nprereg's own §4 declined to do for lack of one on record. That refusal, not\nthe σ machinery, is what made a triple-digit z available here.\n\n## 4. Files and gate\n\n- `research/xchan-at37-02-verify.js`: new — the registered scoring\n arithmetic plus the independent recount, formally embedded\n (`node research/qc/embed.js --streams both --timeout 1800`, 335.1 s,\n code-sha256 `f57889cf…`, out-sha256 `770c0417…`), READINGS appended below\n the banner, binding re-verified by the static `embeds` check.\n- `research/history/staging/xchan-at37-score.md`: this file, HELD for the\n adversarial pass.\n- Nothing else touched: the prereg and the census producer are as this\n session found them; no live document edited; the TODO item X update is\n proposed below, not applied.\n\n**Gate.** Before this session's edits: `node research/qc.js` TOTAL = 0 (the\nstanding census red had cleared when its re-embedded tail landed). After:\nTOTAL = 0, selftest green — see the session record for the runs.\n\n## 5. Proposed TODO item X update (report only, not applied)\n\n> The @37 blind test ran and every registered candidate died (survivor set\n> empty; `xchan-at37-score.md`): measured 1−J = 0.020823 against a nearest\n> registered prediction of 0.021707, a 13.4·S₃ shortfall, independently\n> recounted and confirmed (`research/xchan-at37-02-verify.js`). The\n> finite-level-correction family is CLOSED as the offset's description; the\n> ~3.8 law stands only as a ≈5% scale law, and the \"asymptote exactly 4\"\n> reading is retired in its monotone form. The live question is which\n> structural trend broke: the line bias β's decay or the sieve deficit's\n> 1.15–1.18×4S₂ ratio — undecidable from the net alone. Next instruments, in\n> cost order: (i) the cofactor-one surplus aggregate at @37 in closed form\n> (extend `attack-x-offset-03-cofactor.js`), (ii) the @37 line ensemble\n> (~17–29 h), which separates the split directly.\n\n## 6. Not reached\n\n- **The @37 line ensemble** (β(37)): not run, ~17–29 h; §3's split stays one\n equation in two unknowns until it lands.\n- **The cofactor-one aggregate weight at @37**: a closed-form prime-sum\n computation, proposed above, not performed here.\n- **The full-period recount**: the independent verification covers the full\n @23 period plus 0.404% of the @37 period, contiguously from 0; a defect\n expressed only past 3e10 would evade it (none is known; the 2^53 audit and\n the width-sweep table bound the class).\n- **The CRT denominator by a third path**: it rests on two independent\n embedded producers that agree (the census's pointer tail with its e₃\n identity, and `attack-x-offset-01-terms.js`'s exact sum) — not recomputed\n a third time.\n- **No @41 pricing** and no re-projection of any registered separation.\n", "research/history/staging/xchan-at37-offset-prereg.md": "# Pre-registration: the @37 test of the offset-correction family for the ~3.8 law\n\n\n\n*Written 2026-08-20 and committed alone, before any @37 census exists. The\nproducer instrument already exists — `research/xchan-at29-01-segmented.js` runs\nany level by argument, and `research/attack-x-offset-02-profile.js` shares its\nkernel and gate — but neither has ever been run at @37, no @37 census of any\nkind exists in the corpus, and the level is priced only by extrapolation (\"an\novernight run\", `xchan-at29.md` §7c). Everything below is fixed in advance:\nthe candidate predictions, the σ model and its projection band, the scoring\nrule, and what each verdict does to TODO item X's offset clause.*\n\n## 1. What is being tested\n\n`research/history/staging/xchan-at29.md` left `1 − J = 4S₂` the last law\nstanding and NOT exact: absolute residuals Δ = +1.201e−4 (@29) and +1.185e−4\n(@31), relative offset −0.41% / −0.48%, one detection (z = −2.32 at @31 on the\nslot-clustered σ) and one consistent non-detection (−0.43 at @29).\n`research/attack-x-offset-01-terms.js` (embedded) priced the finite-level\ncorrection family on those residuals; `research/attack-x-offset-02-profile.js`\n(embedded) measured the mechanisms available at the existing levels. The @37\naggregate census decides between the surviving aggregate descriptions.\n\nThe statistic is unchanged and is the one the gate protects: J = obs/CRT with\nobs the mixed super-`W` triple census over the natal set and\nCRT = 6·N̄·Σ_{∏Q>W}1/∏Q, at x = 37, W = 7,420,738,134,810,\nN̄ = 145,286,237,250, K = 198,274 scour primes in (37, 2724079].\n\n## 2. The predictions, all from embedded arithmetic, none adjustable\n\nFrom `research/attack-x-offset-01-terms.js` (its OUTPUT block; sums exact,\ntail mass exact, fitted parameters frozen at their 2026-08-20 values):\n\n| candidate | 1 − J at @37 | mechanism class |\n|---|---|---|\n| **N1** `4S₂` | **0.021863** | no correction: the offset was a fluctuation |\n| **C1** `4S₂ − S₃` | **0.021785** | zero-parameter shrinking correction (POST HOC at @29/@31) |\n| C2 `4S₂ − 2S₃` | 0.021707 | zero-parameter, twice C1's term |\n| M-mult `4S₂(1 − 4.755e−3)` | 0.021759 | constant relative offset |\n| M-abs `4S₂ − 1.186e−4` | 0.021744 | constant absolute residual |\n| M-cS3 `4S₂ − 1.198·S₃` | 0.021770 | fitted S₃ multiple |\n| M-ln `4S₂ − 0.491·S₂/lnW` | 0.021772 | fitted 1/lnW-scale correction |\n\nC4, C8 (≥ 4S₃-size corrections) and the locality shape M-loc are already\nREFUTED at the existing levels and are not carried.\n\n**The error bar, fixed now.** σ_reg = √obs/CRT with CRT(37) = 6.1673e10 exact\ngives σ_reg ≈ 3.98e−6. The slot-clustered inflation is PROJECTED at 2.14,\nbanded [2.12, 2.17] from the measured 2.078 → 2.106 → 2.124 (@23/@29/@31):\n**σ_slot(37) ≈ 8.52e−6**, band [8.44e−6, 8.64e−6]. The run's own printed\nσ_slot supersedes the projection; the projection exists so that the\nseparations below cannot be re-derived after the fact. **The σ estimator is\ncalibrated**: all seven known-truth control draws of\n`research/attack-x-offset-02-profile.js` (random masks at natal density,\ntruth = the level's own J_line) landed inside 1.9σ, so the slot-clustered σ\nneither manufactures nor hides an offset at the scale those draws can see.\n\n**Separations at the projected σ.** N1 vs C1: 9.1σ. C1 vs C2: 9.1σ. C1 vs\nM-abs: 4.8σ. C1 vs M-mult: 3.1σ. C1 vs M-cS3: 1.8σ. C1 vs M-ln: 1.5σ.\nSo @37 decides: (a) whether any correction is needed at all, (b) shrinking\nvs constant-absolute, and (c) C1's coefficient against C2's — but NOT between\nthe members of the shrinking family (C1, M-cS3, M-ln, and M-mult only weakly),\nand no claim of that kind may be written from @37 alone.\n\n## 3. The scoring rule, fixed now\n\nCompute z_c = ((1−J)_meas − pred_c)/σ_slot for every candidate, on the run's\nown slot-clustered σ_slot.\n\n- A candidate is **KILLED** by |z_c| > 3 and **ALIVE** otherwise; the family\n verdict names every survivor, sorted by |z_c|.\n- **If N1 survives** (|z_N1| ≤ 3), the offset's status reverts to \"not\n established as structure\": two sharp levels plus a third failing to\n reproduce it means the @31 detection does not generalize, and TODO item X's\n offset clause must say so.\n- **If N1 is killed and at least one shrinking-family member survives**, the\n offset is CONFIRMED as structure with a shrinking scale; C1's status\n upgrades from POST HOC to \"survived one blind level\" only if |z_C1| ≤ 3.\n- **If only M-abs survives**, the residual is a constant-absolute anomaly, no\n candidate mechanism on file produces it, and the item must carry that.\n- **J-band clause** (unchanged from `xchan-at29-prereg.md`): J(37) outside\n (0.94, 1.00] means the object changed; the report leads with that.\n- **Gate**: the producer must reproduce its own embedded reference levels\n before @37 is reported (the same gate the @29/@31 run used); a gate failure\n reports an instrument failure and no @37 value.\n\n## 4. Secondary, registered but not decisive\n\n- **The line's lattice bias.** `attack-x-offset-02-profile.js` measured\n β = J_line − 1 = +9.88e−3, +7.44e−3, +5.38e−3, +4.03e−3 over @19..@31\n (z = 2.3, 8.7, 34.7, 147.6): positive, decaying, and 16–24% of the natal\n deficit's own size. If the @37 run also runs the line ensemble (roughly\n 3–5× the natal cost, so it may be deferred), the registered reading is the\n continuation of that decay and of 1 − J/J_line (1.18× and 1.15× 4S₂ at\n @29/@31); no numeric band is fixed because no β model is on record. The\n sighting β/S₃ ≈ 39–43 over the four levels is recorded here so that, if a\n β model is ever fitted, this note is its provenance.\n- **P1**: aligned super-`W` obs = 0 at @37, an instrument check as always.\n\n## 5. Scope\n\nProducer: `research/xchan-at29-01-segmented.js -- ... 37` or the profile\ninstrument at @37; either must be embedded via `node research/qc/embed.js`\nwith its gate green. Report: an extension of\n`research/history/staging/item-x-offset.md` or a successor staging note. No\nlive document is edited by the @37 run; TODO item X's update is proposed in\nthe report, not applied. This file is committed alone before any @37 number\nof any kind exists.\n"}, "offline_replay_output": "served research/QUESTIONS.md sha256(LF) e2ddcfc55f89be535041515a26fbc5068db832a7db548d35765b75f5360ff60e\nserved /questions: 217 questions, 53 open or partial\nsweeping rows 38..52: 15 rows, note set below\n\nROW 38 Q-rec-cheapest-0830 PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\nROW 39 Q-recon-0904-sifting-limit-floor PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\nROW 40 Q-rho-maximal-law PARTIAL status_ok=True verdict_ok=True -> CURRENT (2/2 note(s) with a ledger block)\nROW 41 Q-rho-sup-z41-0830 PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\nROW 42 Q-rml-proof-0829n PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\nROW 43 Q-fekete-1d-defect47 PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\nROW 44 Q-g2-falls-decision-rule PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\nROW 45 Q-hsub-reductions PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\nROW 46 Q-xchannel-at23 PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\nROW 47 Q-xchannel-closedform PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\nROW 48 Q-xchannel-offset PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\nROW 49 Q-at43-bigint-0830 PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\nROW 50 Q-comb-discrepancy PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\nROW 51 Q-comb-tail-0830 PARTIAL status_ok=True verdict_ok=True -> CURRENT (1/1 note(s) with a ledger block)\nROW 52 Q-c2prime-drift PARTIAL status_ok=True verdict_ok=True -> CURRENT (2/2 note(s) with a ledger block)\n\n15 rows swept, 0 stale at the registry level.\nSilent on whether a ledger block is itself stale against its note body,\nriders and the outcome register: see rows23-37.md for that column.\n"}