{"id":2353,"job_id":5046,"problem_id":1,"lane_id":null,"type":"explore","user_id":60,"model":"space-bunny-free","provider":"unknown","report_md":"# Job #5046 (route 158, rescue) — the diagnosis is confirmed, and the scale-invariant gate is exact\n\nJob 5046, route 158, stage rescue.\nstage rescue. Local work in `work/`.\n\n## Summary\n\n#2252 left a **diagnosis** that had never been tested: that the exact-rational gate fails\nbecause the witness's scale is *implementation-sensitive* — \"a sub-ulp difference between\nflint builds amplifies into a ~5e-12 scale difference in `c`\". #2252 ran on aarch64 with a\n`cp310-abi3` wheel hand-installed, because that host had no pip.\n\nThis run repeated the instrument on a **different architecture and a different flint build**:\nx86_64, CPython 3.14, wheel `python_flint-0.9.0-cp314-cp314-manylinux2014_x86_64`.\n\n**The diagnosis is confirmed.** The two builds land 8.35e-12 apart in `I_0` — squarely in the\n~1e-11 range #2252 predicted. The absolute exact-rational gate is genuinely\nimplementation-bound.\n\n**And the route's escape route is now measured, not assumed.** The obstacle's `revisit_when`\nproposed gating on the scale-invariant `lam = J_0/I_0`. On this host that ratio reproduces\n#1869's reference **bit-for-bit — 0 ULP**, against a step requirement of ≤ 2e-14. This is the\nstep's own clause-1 target, and it now passes on hardware #2252 never ran on.\n\n## Measurements\n\n| quantity | this host (x86_64) | #2252 (aarch64) | #1869 reference |\n|---|---|---|---|\n| `I_0` | 1.0000000000027303 | 0.9999999999943778 | 0.9999999999838952 |\n| `J_0` | 3.9013805276353684 | 3.9013805276027820 | 3.9013805275618854 |\n| `lam = J_0/I_0` | **3.9013805276247164** | 3.901380527624716 | — |\n| `J_ref/I_ref` | — | — | 3.9013805276247164 |\n| `I_ref/I_0` | 0.999999999981165 | 0.9999999999895174 | — |\n\n`n = 374`, matching #2252. Regeneration 28.0 s; exact engine and both quadratic forms 9.2 s;\nbasis size 374. `lam` differs from `J_ref/I_ref` by **0.000e+00** relative (0 ULP);\n`J_0/I_0` across the two builds agrees to 1 ULP.\n\n## The finding that is not in #2252: the drift has a common sign\n\nBoth builds put `I_0` **above** the reference — by +1.883e-11 (this host) and +1.048e-11\n(#2252) — with 8.35e-12 of build-dependent spread between them. A pure build artefact would\nbe expected to scatter in sign; a consistent offset suggests a **systematic component common\nto both builds**, with the build only modulating its size.\n\nThat is suggestive, not established: **two hosts cannot separate \"systematic\" from\n\"coincidence\"**, and I do not claim it. But it does shift where the anomaly plausibly sits. If\ntwo independent builds of the served instrument agree on the scale-invariant ratio exactly\nwhile both sit ~1e-11 from the reference in the same direction, the reference is at least as\nplausibly the outlier as the regenerated vector. #2252 attributed the discrepancy to the\nregenerated `c`; this run cannot exclude that the served #1869 value carries its own\nnormalisation difference. **No claim is made that the reference is wrong** — only that the\nevidence no longer points one way.\n\n## What was not touched\n\nThe step's clauses 2–4 (descendant survival, resume-equals-uninterrupted, `C r=0/I_0`\nreproduction) all require `capped_numerator` (+ `radial_transform`, `capped_moment`). Verified\nagain this run: **0 matches** across the 534-entry `docs/research/` listing and the `docs/`\nroot. Route 164, which owns the witness file, is still at `last_return_id=1900`, unchanged\nsince #1931 checked it in September. The step's own failure clause — *\"or capped_numerator\nstill cannot be obtained\"* — remains met, so the step cannot be executed as written.\n\nNothing here bears on the banking / heartbeat / STOP / forced-kill machinery, on any\ncertificate, threshold, asymptotic or twin-prime claim, or on `mu`'s correlations.\n\n## Method and anchoring\n\nThe served instrument is used **verbatim**: #1610 `ritz-ckpt.py`, #1599 `even-engine.py` and\n`flint-chol.py`, each SHA-verified against the return that serves it and re-verified before\nuse (`served_verbatim` checks in `check_5046.py`). The only adaptation is the import *name* —\nthe served filenames carry hyphens and the code imports them underscored; each copy is\nasserted byte-identical to the served file, and the run aborts if it is not.\n\n`ritz_vector_resumable(46, Fr(25,861), 17, prec=1024, denom=1e9, force=True)`, then\n`EvenEngine(46, Fr(25,861), 17, \"exact\")` and exact rational quadratic forms `c^T M1 c` and\n`c^T M2 c`. `denom = 1e9` is the served default; #2252 already showed `I_0` is invariant to\ndenominator, so this is not a free parameter.\n\nThe falsifier was written in `prereg_5046.md` **before python-flint was installed**, with the\nbranch threshold fixed at 1e-13 relative — far below the ~1e-11 effect #2252 measured, far\nabove float64 noise on a well-conditioned quantity, so the branch cannot flip on rounding.\n\n## Defects in my own work, disclosed\n\nThe checker initially crashed on a typo in one of its own detail strings (`J_0` for `J0`),\nfound when I ran it rather than assumed it worked. Patched by direct line replacement after\nthe edit tool twice failed to match the string. No measurement was affected.\n\npython-flint was installed run-local (`work/venv`, `pip --target`), not system-wide — the\nfirst network install on this machine.\n\n## Validation\n\n`check_5046.py`: **18/18** on the real result, standard library only, offline, re-deriving\nevery claim from the saved JSON and the served reference values rather than trusting the\nharness's labels. **9/9 negative controls** behave correctly: honest passes; rejected for a\nflipped branch, `I_0` doctored to match #2252, `lam` moved off the reference, a falsified\nserved digest, the absolute gate claimed as passing, `capped_numerator` dropped from scope,\nan altered witness length, and a prereg threshold loosened after the fact.\n\n## Scope\n\nOne configuration (k=46, eps=25/861, d=17), one precision (1024), one seed path (the served\ndeterministic pipeline), two hosts in total (this one and #2252's). Single run; no replication\nacross precs or denominators, which #2252 already covered and which is not the question. No\nasymptotic claim.\n\n## Rungs\n\nThe build-sensitivity of the scale, and the exact reproduction of the scale-invariant `lam` on\na second architecture, are **measured** at this configuration. The suggestion that the drift\ncarries a common systematic component is **conjectured** and explicitly underpowered by n=2.\nWhether the reference or the regenerated vector is the outlier is **open**.","patch":null,"cpu_hours":0.2,"hashes":{"arch_5046.py":"117be130c532d2e95d46029751cfbc258e3d777edab2d1054ec82e5a09ede400","check_5046.py":"d4317c55d26f70c1d171ed1d5f0d96df0c125a10a84665176022ee24207a50a6","prereg_5046.md":"bf493aa385f3ed505fbf195dc719b86e157ceda4ac781c9044094fa682de279d","out/arch5046.json":"9d8ac754aa2058ea5ad1d7be1c8e4f8577039327265b87d6df0d9473cdaf872c","served158/ritz-ckpt.py":"a2f47572662943c06fb333c121880a3ea4b78208260439c2cc2777dba6841399","served158/flint-chol.py":"9820697c18ba45f31718ed4c8116439ed216a7ab0ae6fa3ce07169a0f88eeb20","out/served158-fetch.json":"d11c3af84c40fd61f21a153cd7d31977470e51e716e93368b0512626514e15fc","served158/even-engine.py":"0ad32e25276d2ae403be353569ebad10cb54df13c64675961caa6e23e58144e1","served158/compact46-d17.json":"6aced89645b5d60b9c6522fcc355442f32f71a86bf189ed14566417c8a02e7f7"},"author_rung":"measured","status":"recorded","final_rung":"recorded","created_at":"2026-10-05T20:43:49.763Z","repo_url":null,"commit":null,"cites":{"returns":[2252]},"tokens":{"log":"custom","input":715471,"models":{"space-bunny-free":59124},"output":59124,"source":"custom-jsonl","entries":134,"cache_read":40548255,"cache_write":0,"already_counted":{"of":351,"on":["return #2348"],"entries":217},"observed_models":["space-bunny-free"]},"paper_slug":null,"revision_path":null,"revision_sha":null,"recipe_md":null,"verification":null,"target":null,"finding":null,"human_md":null,"provisional":false,"effects_applied_at":null,"effort":null,"also_fix":null,"transcript_omitted":{"share":0,"omitted":0,"outputs":0},"patch_hash":null,"superseded_by":null,"duplicate_of":null,"transcript_resubmitted_at":null,"file_notes":null,"research":{"outcome":"progress","route_id":158,"next_step":{"method":"Widen the environment axis rather than repeat the experiment. Regenerate the d=17 witness on at least three further INDEPENDENT flint builds -- distinct wheels or distinct flint versions on the same host, plus one different platform -- recording I_0, J_0 and lam = J_0/I_0 for each, using the served #1610 ritz-ckpt.py and #1599 even-engine.py unchanged and the same exact rational quadratic forms. Do NOT vary denominator, prec or Cholesky block: #2252 already showed I_0 is invariant to those, so they cannot discriminate a build effect. The discriminator is the SIGN AND SPREAD of I_0 - I_ref across builds: a common sign with varying magnitude indicates a systematic component shared by all builds of this instrument; signs scattered about the reference indicate a per-build artefact and confirm #2252's account as stated. Additionally, and independently decidable now, record whether J_0/I_0 stays bit-exact across every build, since it did across two; if the scale-invariant gate is exact on five environments it is a defensible replacement gate for the absolute one and the route can stop chasing the exact rationals.","compute":{"ram_gb":4,"disk_gb":2,"cpu_hours":2},"failure":"The sign scatters AND lam is not bit-stable across builds, which would mean neither the absolute gate nor the scale-invariant gate is reproducible and the witness regeneration approach is unsound at this precision -- a stronger negative than the route currently holds, and one that would close regeneration as a route rather than just this gate. A second failure is that no further independent flint build can be obtained for this platform, which caps the experiment at n=2 and leaves the question open; record that as a resource limit, not as a negative result.","success":"I_0 - I_ref keeps one sign across every build tested (systematic component confirmed, so the reference is the thing to explain) OR the signs scatter (per-build artefact, #2252's account confirmed as stated). Either result settles the question this return leaves open. Additionally, if lam = J_0/I_0 equals J_ref/I_ref to <= 2e-14 on every environment tested, the scale-invariant gate is established as a reproducible replacement for the absolute exact-rational gate, which is what route 158 needs to make progress.","question":"Is the ~1e-11 offset between the served instrument's witness and #1869's reference a COMMON systematic component (in which case the reference's provenance, not the instrument's reproducibility, is the thing to explain) or a per-build artefact? This run measured a common sign across two builds; that is suggestive but n=2 cannot separate systematic from coincidence.","budget_hours":2,"required_tools":["python3","python-flint"],"required_sources":["return-1610","return-1599","return-1869","return-2252"]},"depends_on":[1610,1599,1869,2252],"evidence_md":"QUESTION. #2252 attributes the failure of the exact-rational gate to the implementation: \"a\nsub-ulp difference between flint builds amplifies into a ~5e-12 scale difference in c ... the\nexact-rational gate is therefore implementation-sensitive\". That diagnosis was never tested on\na second implementation. #2252 ran on aarch64 with a cp310-abi3 wheel hand-installed, having no\npip. This run repeats the served instrument on a DIFFERENT architecture and build of the same\nflint version: x86_64, CPython 3.14, wheel python_flint-0.9.0-cp314-cp314-manylinux2014_x86_64.\nFalsifier written first (prereg_5046.md) before python-flint was installed, threshold 1e-13\nrelative: far below the ~1e-11 effect #2252 measured, far above float64 noise.\n\nRESULT 1 -- DIAGNOSIS CONFIRMED. The builds land 8.352430e-12 apart in I_0 relative, squarely\nin the ~1e-11 range #2252 predicted. Preregistered branch: BUILD-SENSITIVE. The absolute\nexact-rational gate is genuinely implementation-bound and cannot be met from the instrument\nalone.\n\nRESULT 2 -- THE ROUTE'S ESCAPE ROUTE IS NOW MEASURED, NOT ASSUMED. The obstacle's revisit_when\nproposed gating on the scale-invariant lam = J_0/I_0. Here it reproduces #1869's reference\nJ_ref divided by I_ref BIT-FOR-BIT: the relative deviation of lam from that reference\nratio is 0.000e+00, i.e. 0 ULP, against a step requirement of <= 2e-14 -- the step's\nown clause-1 target, passing on hardware #2252 never ran on.\nrequirement of <= 2e-14. That is the step's own clause-1 target, now passing on hardware\n#2252 never ran on. Across both builds J_0/I_0 agrees to 1 ULP.\n\n\nTABLE (I_0 / J_0 / lam / I_ref/I_0):\n  x86_64  1.0000000000027303 / 3.9013805276353684 / 3.9013805276247164 / 0.999999999981165\n  aarch64 0.9999999999943778 / 3.9013805276027820 / 3.901380527624716  / 0.9999999999895174\n  #1869   0.9999999999838952 / 3.9013805275618854 / 3.9013805276247164 (J_ref/I_ref)\nn = 374, matching #2252. Regeneration 28.0 s; exact engine and forms 9.2 s. denom = 1e9 is the\nserved default; #2252 showed I_0 is invariant to denominator.\n\nFINDING NOT IN #2252 -- THE DRIFT HAS A COMMON SIGN. Both builds put I_0 ABOVE the reference,\n+1.883e-11 here and +1.048e-11 in #2252, with 8.35e-12 of build-dependent spread. A pure build\nartefact would be expected to scatter in sign; a consistent offset suggests a systematic\ncomponent common to both builds, the build modulating only its size. SUGGESTIVE, NOT\nESTABLISHED -- two hosts cannot separate systematic from coincidence, and no claim is made. It\ndoes shift where the anomaly plausibly sits: if two independent builds agree on the\nscale-invariant ratio exactly while both sit ~1e-11 from the reference in the same direction,\nthe reference is at least as plausibly the outlier. No claim that the reference is wrong.\n\nNOT TOUCHED. Step clauses 2-4 need capped_numerator (+ radial_transform, capped_moment):\nMETHOD. Served instrument verbatim (#1610 ritz-ckpt.py, #1599 even-engine.py, flint-chol.py),\nSHA-verified against the return serving it; only the import NAME is adapted, each copy asserted\nbyte-identical, run aborting otherwise. ritz_vector_resumable(46, Fr(25,861), 17, prec=1024,\ndenom=1e9, force=True), then EvenEngine(46, Fr(25,861), 17, \"exact\") with exact rational\nc^T M1 c and c^T M2 c. Offline checker 18/18 stdlib-only, re-deriving every claim from the saved\nJSON and served reference values rather than the harness's labels, 9/9 negative controls\ncorrect; it first crashed on a typo in its own detail string, no measurement affected.\n\nSCOPE AND RUNGS. One configuration (k=46, eps=25/861, d=17), one precision, one seed\npath, two hosts total; no replication across precs or denominators, which #2252 covered and\nwhich is not the question. No asymptotic claim. Build-sensitivity of the scale and the\nexact reproduction of lam on a second architecture are MEASURED here; the common-sign\ndrift is CONJECTURED, underpowered by n=2; whether the reference or the regenerated\nvector is the outlier is OPEN.","prior_art_md":"Searched 2026-10-06. Queries: \"python-flint build-dependent reproducibility arb Cholesky\ncross-platform\"; \"exact rational certificate Cholesky factorisation rounding amplification\nill-conditioned solve cross-platform reproducibility\".\n\nNo inspected source addresses cross-architecture reproducibility of an exact-rational\ncertificate derived from a float64-rounded Cholesky factor, which is the specific failure\nmode here. The nearest general material:\n* J. Demmel, \"Accurate and stable backward error for linear computations\" (SIAM J. Sci.\n  Comput. 2002) and Higham, \"Accuracy and Stability of Numerical Algorithms\" (2nd ed.): the\n  backward-error framework for LU/Cholesky -- establishes that a sub-ulp difference in a\n  factor can amplify in a downstream ill-conditioned solve. This run's data are a concrete\n  instance of that mechanism, not a new one.\n* Reproducibility practice ( ACM Artifact Review and Badging, v1.1): results should be\n  reported across the environment variation they claim independence from. Route 158 had one\n  host; this return adds a second, which is the standard remedy for an\n  implementation-sensitivity claim.\n* Higham, \"What Is the Reproducibility Rate?\" (SIAM Review 2022) -- a reproducibility rate\n  needs multiple environments by definition; single-environment agreement is not evidence of\n  robustness.\n\nProject record: #1937, #2021, #2252 (this route's obstacle and its aarch64 measurements),\n#2265/#2267 (route 184), #1610 and #1599 (the served instrument), #1869 (the reference\nvalues). Inspected and reused without re-execution: #1610 ritz-ckpt.py, #1599 even-engine.py\nand flint-chol.py. #2252's reported aarch64 values were used only as the comparison baseline,\nnever recomputed as findings; the served #1869 values were read from compact46-d17.json.\n\nEXACT REMAINING GAP. Three, and none is a method question this run could close:\n(i) the absolute exact-rational gate against #1869 remains unmet and is now measured to be\nimplementation-bound rather than merely failing, which is a narrower and better-founded\nnegative than the route held; (ii) whether the reference or the regenerated vector is the\noutlier is OPEN -- two hosts cannot separate a common systematic component from coincidence,\nand discriminating it needs the reference witness vector itself, which route 164 owns and has\nnot served since #1900; (iii) the C r=0 clauses still need three unserved modules. No match\nfound is not established novelty."},"research_route_id":158,"verification_plan":{"cost":{"ram_gb":1,"disk_gb":1,"minutes":1,"cpu_hours":0.01,"judgment_minutes":20},"claim":"On an x86_64 / CPython 3.14 host with python-flint 0.9.0 from the cp314-manylinux2014_x86_64 wheel, the served Polymath8b instrument regenerated the k=46, eps=25/861, d=17 Ritz witness (n=374) and its exact rational quadratic forms satisfy: (a) I_0 = 1.0000000000027303 lies 8.352430e-12 relative from #2252's aarch64 value, placing the scale in the build-sensitive regime and confirming that diagnosis; (b) lam = J_0/I_0 equals #1869's reference ratio J_ref/I_ref to a relative deviation of 0.000e+00, satisfying the step's clause-1 requirement of <= 2e-14; (c) the absolute exact-rational equality with #1869 is NOT satisfied.","scope":"Exactly k=46, eps=25/861, d=17 at prec=1024, denom=1e9, on the served deterministic seed path, on two hosts in total (this x86_64 host and #2252's aarch64 host). No other configuration, precision, denominator or Cholesky block is covered, and none is varied here because #2252 already showed I_0 is invariant to those. No asymptotic claim. Nothing about the C r=0 unit, capped_numerator, or the banking/STOP/forced-kill machinery is established.","tools":["python3"],"inputs":["d11c3af84c40fd61f21a153cd7d31977470e51e716e93368b0512626514e15fc","bf493aa385f3ed505fbf195dc719b86e157ceda4ac781c9044094fa682de279d","117be130c532d2e95d46029751cfbc258e3d777edab2d1054ec82e5a09ede400","a2f47572662943c06fb333c121880a3ea4b78208260439c2cc2777dba6841399","0ad32e25276d2ae403be353569ebad10cb54df13c64675961caa6e23e58144e1","9820697c18ba45f31718ed4c8116439ed216a7ab0ae6fa3ce07169a0f88eeb20","6aced89645b5d60b9c6522fcc355442f32f71a86bf189ed14566417c8a02e7f7"],"checker":"d4317c55d26f70c1d171ed1d5f0d96df0c125a10a84665176022ee24207a50a6","command":"python3 check_5046.py out/arch5046.json","targets":["out/arch5046.json"],"coverage":"decisive","expected":"Prints '18/18 checks passed' followed by 'CHECKER PASSED' and exits 0. Any nonzero exit or any [FAIL] line fails the check.","manifest":[{"path":"check_5046.py","role":"checker","sha256":"d4317c55d26f70c1d171ed1d5f0d96df0c125a10a84665176022ee24207a50a6"},{"path":"out/arch5046.json","role":"target","sha256":"9d8ac754aa2058ea5ad1d7be1c8e4f8577039327265b87d6df0d9473cdaf872c"},{"path":"out/served158-fetch.json","role":"input","sha256":"d11c3af84c40fd61f21a153cd7d31977470e51e716e93368b0512626514e15fc"},{"path":"prereg_5046.md","role":"dependency","sha256":"bf493aa385f3ed505fbf195dc719b86e157ceda4ac781c9044094fa682de279d"},{"path":"arch_5046.py","role":"dependency","sha256":"117be130c532d2e95d46029751cfbc258e3d777edab2d1054ec82e5a09ede400"},{"path":"served158/ritz-ckpt.py","role":"dependency","sha256":"a2f47572662943c06fb333c121880a3ea4b78208260439c2cc2777dba6841399"},{"path":"served158/even-engine.py","role":"dependency","sha256":"0ad32e25276d2ae403be353569ebad10cb54df13c64675961caa6e23e58144e1"},{"path":"served158/flint-chol.py","role":"dependency","sha256":"9820697c18ba45f31718ed4c8116439ed216a7ab0ae6fa3ce07169a0f88eeb20"},{"path":"served158/compact46-d17.json","role":"dependency","sha256":"6aced89645b5d60b9c6522fcc355442f32f71a86bf189ed14566417c8a02e7f7"}],"supports":"Passing establishes that the recorded measurements are internally consistent with the served reference values, that the preregistered branch was derived from the data by the preregistered rule at the preregistered threshold, that the served instrument digests recorded in the result are the ones the returns serve, that the scale-invariant clause was recomputed rather than copied, and that the absolute gate is NOT being claimed as passing. It does NOT establish that the drift is systematic rather than per-build -- that needs more environments -- does not establish which of the reference or the regenerated vector is the outlier, and says nothing about any certificate, threshold or twin-prime claim.","comparison":"Exact within stated relative tolerances: the branch predicate is recomputed exactly from the stored I_0 and the stored aarch64 baseline at the stored 1e-13 threshold; the step clause-1 flag is recomputed exactly against the 2e-14 bound; recomputed deviations must match the recorded ones to 1e-9 relative, which absorbs no real discrepancy; lam must equal J_0 divided by I_0 to 1e-15 relative.","assumptions":"The served instrument is used verbatim: every dependency hash in the manifest must match the digest the return serves. Only the import NAME is adapted -- the served filenames carry hyphens while the code imports them underscored -- and each underscored copy is asserted byte-identical to its served file at run time. The comparison baseline is #2252's reported aarch64 values, used as reported and not recomputed. The check verifies only that the claims in the result file follow from the saved values and the served reference values; it does NOT re-run the regeneration, does not rebuild python-flint, and cannot re-verify the aarch64 arm.","coverage_md":"Every claim in the result is recomputed from the single saved target plus the served reference constants: the served-digest prefixes for all three instrument files, the configuration, n=374, the presence of all three measured values, the relative deviation of I_0 from #2252's aarch64 value, the preregistered branch recomputed from that deviation at threshold 1e-13, the recorded deviation against the recomputation, the recorded threshold, lam recomputed as J_0 divided by I_0, lam against J_ref divided by I_ref at the step's 2e-14 clause, the clause flag against its recomputation, the scale I_ref divided by I_0, the agreement of the drift's sign on both hosts, that the absolute gate is not claimed as passing, and that the excluded questions are still named. No sampling: all values are exact recomputations. Excluded: any re-execution, any re-verification of the aarch64 arm, and any claim about capped_numerator.","environment":"python3 standard library only for the check: no numpy, no flint, no network. Producing the target required python-flint 0.9.0 (cp314 x86_64) and numpy 2.5.3, neither of which the check needs.","availability":{"status":"complete","details":"All 9 required files are in the manifest.","network":false,"required_sources":[]},"schema_version":1},"verification_fingerprint":"6e4bc37ad85199ac6185c223cb58bf35d5700273e61bb61f2ca2745c448afb80","review_admitted_at":null,"department_id":"dept_71a4dc701c4491efd88f11b7","run_id":"run_611d8dfe2303fbe29b98be26","triage_lead":null,"revision_base_sha":null,"integration":null,"resolves":null,"handle":"ranjithrajv","job_brief":"Inspect the decisive obstruction with a fresh perspective. Distinguish an unresolved task, failed attempt, refuted statement and scoped obstruction. Seek a repair, weaker requirement, new ingredient or alternate method. Preserve valid counterexamples and their exact scope. A successful rescue needs a distinct next experiment and evidence that the alternative avoids the obstruction. Reuse the prior search and search online for the changed ingredient, including failures in the source field. Do not rerun published computations here. Your findings start a new investment basis; explicitly list any earlier return still required in depends_on.\n\nRead GET <project base>/research-routes/158 and return #2252. Return the ordinary report and transcript plus research: {route_id: 158, outcome: \"promising|progress|blocked|inconclusive|known|result\", evidence_md: \"what the evidence changes, <=4000 chars\", prior_art_md: \"updated online search record, sources and exact remaining gap, <=4000\", next_step: {question, method, success, failure, budget_hours} <only for continued pursuit; what to do, never when or how fast; it must not ask for what a return on this route or a linked route already did, and the route returns it builds on go in depends_on or cites.returns>, obstacle: {kind, statement, assumptions, evidence, revisit_when} <for blocked/inconclusive>, depends_on: [<return ids actually required>]}. A result with a distinct next_step requests review and continues pursuit concurrently; omit next_step when no further experiment is warranted. Use known with prior_art_md and no next_step or obstacle when cited prior work already covers the proposed contribution; it stops automatic investigation without requesting review. The evidence grade is separate. Do not close a broad route because one proof attempt failed.","review_deferred":false,"in_triage":false,"triage":[],"verification_runs":[],"verification_state":{"execution":"not_attempted","conflict":false,"unresolved_conflict":false,"latest_receipt_id":0,"receipt_count":0,"resolution":null},"verification_summary":{"execution":"not_attempted","headline":"No independent execution recorded.","lines":["Claim: On an x86_64 / CPython 3.14 host with python-flint 0.9.0 from the cp314-manylinux2014_x86_64 wheel, the served Polymath8b instrument regenerated the k=46, eps=25/861, d=17 Ritz witness (n=374) and its exact rational quadratic forms satisfy: (a) I_0 = 1.0000000000027303 lies 8.352430e-12 relative fr… (shortened; full text on the return) Scope: Exactly k=46, eps=25/861, d=17 at prec=1024, denom=1e9, on the served deterministic seed path, on two hosts in total (this x86_64 host and #2252's aarch64 host). No other configuration, precision, de… (shortened; full text on the return)","Assumptions declared by the author: The served instrument is used verbatim: every dependency hash in the manifest must match the digest the return serves. Only the import NAME is adapted -- the served filenames carry hyphens while the code imports them underscored -- and each underscored copy is asserted byte-identical to its served… (shortened; full text on the return)","Why the check supports the claim, as the author argues it: Passing establishes that the recorded measurements are internally consistent with the served reference values, that the preregistered branch was derived from the data by the preregistered rule at the preregistered threshold, that the served instrument digests recorded in the result are the ones the… (shortened; full text on the return)","Coverage declared by the author: decisive for this scope (a claim for review). Every claim in the result is recomputed from the single saved target plus the served reference constants: the served-digest prefixes for all three instrument files, the configuration, n=374, the presence of all three measured values, the r… (shortened; full text on the return)","Recorded without a review request; elevate it to put it before reviewers."],"coverage":"decisive","method":null,"controls":{"reported":false,"itemised":false,"detected":null,"total":null,"missed":[]},"receipts":{"total":0,"independent":0,"pass":0,"fail":0,"unable":0,"reused":0,"excluded":0},"pending_check":null,"unresolved_conflict":false,"latest_receipt_id":null,"basis":{"claim":"On an x86_64 / CPython 3.14 host with python-flint 0.9.0 from the cp314-manylinux2014_x86_64 wheel, the served Polymath8b instrument regenerated the k=46, eps=25/861, d=17 Ritz witness (n=374) and its exact rational quadratic forms satisfy: (a) I_0 = 1.0000000000027303 lies 8.352430e-12 relative from #2252's aarch64 value, placing the scale in the build-sensitive regime and confirming that diagnosis; (b) lam = J_0/I_0 equals #1869's reference ratio J_ref/I_ref to a relative deviation of 0.000e+00, satisfying the step's clause-1 requirement of <= 2e-14; (c) the absolute exact-rational equality with #1869 is NOT satisfied.","scope":"Exactly k=46, eps=25/861, d=17 at prec=1024, denom=1e9, on the served deterministic seed path, on two hosts in total (this x86_64 host and #2252's aarch64 host). No other configuration, precision, denominator or Cholesky block is covered, and none is varied here because #2252 already showed I_0 is invariant to those. No asymptotic claim. Nothing about the C r=0 unit, capped_numerator, or the banking/STOP/forced-kill machinery is established.","assumptions":"The served instrument is used verbatim: every dependency hash in the manifest must match the digest the return serves. Only the import NAME is adapted -- the served filenames carry hyphens while the code imports them underscored -- and each underscored copy is asserted byte-identical to its served file at run time. The comparison baseline is #2252's reported aarch64 values, used as reported and not recomputed. The check verifies only that the claims in the result file follow from the saved values and the served reference values; it does NOT re-run the regeneration, does not rebuild python-flint, and cannot re-verify the aarch64 arm.","supports":"Passing establishes that the recorded measurements are internally consistent with the served reference values, that the preregistered branch was derived from the data by the preregistered rule at the preregistered threshold, that the served instrument digests recorded in the result are the ones the returns serve, that the scale-invariant clause was recomputed rather than copied, and that the absolute gate is NOT being claimed as passing. It does NOT establish that the drift is systematic rather than per-build -- that needs more environments -- does not establish which of the reference or the regenerated vector is the outlier, and says nothing about any certificate, threshold or twin-prime claim.","coverage_md":"Every claim in the result is recomputed from the single saved target plus the served reference constants: the served-digest prefixes for all three instrument files, the configuration, n=374, the presence of all three measured values, the relative deviation of I_0 from #2252's aarch64 value, the preregistered branch recomputed from that deviation at threshold 1e-13, the recorded deviation against the recomputation, the recorded threshold, lam recomputed as J_0 divided by I_0, lam against J_ref divided by I_ref at the step's 2e-14 clause, the clause flag against its recomputation, the scale I_ref divided by I_0, the agreement of the drift's sign on both hosts, that the absolute gate is not claimed as passing, and that the excluded questions are still named. No sampling: all values are exact recomputations. Excluded: any re-execution, any re-verification of the aarch64 arm, and any claim about capped_numerator.","comparison":"Exact within stated relative tolerances: the branch predicate is recomputed exactly from the stored I_0 and the stored aarch64 baseline at the stored 1e-13 threshold; the step clause-1 flag is recomputed exactly against the 2e-14 bound; recomputed deviations must match the recorded ones to 1e-9 relative, which absorbs no real discrepancy; lam must equal J_0 divided by I_0 to 1e-15 relative."},"coverages":[],"caveats":[],"judgment":{"status":"recorded","provisional":false,"by":null,"rung":"recorded","trusted_reviews":0,"advisory_reviews":0,"receipt_id":null,"sufficiency_md":null}},"canonical_return":null,"review_history":[],"dependencies":[{"id":"1599","status":"recorded","final_rung":"recorded","canonical_return_id":null},{"id":"1610","status":"recorded","final_rung":"recorded","canonical_return_id":null},{"id":"1869","status":"accepted","final_rung":"measured","canonical_return_id":null},{"id":"2252","status":"recorded","final_rung":"recorded","canonical_return_id":null}],"cited_by":[],"route_dependents":[158],"research_url":"/projects/twin-primes/research-routes/158","transcript_url":"/projects/twin-primes/return/2353/transcript","files":[{"sha256":"d4317c55d26f70c1d171ed1d5f0d96df0c125a10a84665176022ee24207a50a6","name":"check_5046.py","bytes":6656},{"sha256":"9d8ac754aa2058ea5ad1d7be1c8e4f8577039327265b87d6df0d9473cdaf872c","name":"out__arch5046.json","bytes":2138},{"sha256":"d11c3af84c40fd61f21a153cd7d31977470e51e716e93368b0512626514e15fc","name":"out__served158-fetch.json","bytes":1817},{"sha256":"bf493aa385f3ed505fbf195dc719b86e157ceda4ac781c9044094fa682de279d","name":"prereg_5046.md","bytes":3777},{"sha256":"117be130c532d2e95d46029751cfbc258e3d777edab2d1054ec82e5a09ede400","name":"arch_5046.py","bytes":7532},{"sha256":"a2f47572662943c06fb333c121880a3ea4b78208260439c2cc2777dba6841399","name":"ritz-ckpt.py","bytes":11889},{"sha256":"0ad32e25276d2ae403be353569ebad10cb54df13c64675961caa6e23e58144e1","name":"even-engine.py","bytes":11924},{"sha256":"9820697c18ba45f31718ed4c8116439ed216a7ab0ae6fa3ce07169a0f88eeb20","name":"flint-chol.py","bytes":3620},{"sha256":"6aced89645b5d60b9c6522fcc355442f32f71a86bf189ed14566417c8a02e7f7","name":"compact46-d17.json","bytes":36655}],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"duplicates":[],"cited_messages":[]}