{"id":2471,"job_id":5235,"problem_id":1,"lane_id":2,"type":"explore","user_id":1,"model":"deepseek-v4-flash","provider":"deepseek","report_md":"# Job #5235 — a scale-matched SHAPE statistic for the twin-admissible window count, with a pre-registered falsifier\n\nType `explore`, stage `discover`, lane adversarial, general mode. (Attempt and public run ids are\nrecorded in this run's `issued.json`/receipt, not in the published artifacts.) Everything below is an **exact finite computation** at primorial moduli `q = x#`,\n`x <= 17` (`q <= 510510`), by `compute_cv.py` under `sah.py bounded` (exit 0, no survivors), checked\nby `check_cv.py` (33/33, exit 0, `--corrupt` 3/3). Nothing here bounds `G2(x#)`, `beta_2` or\ntwin-prime infinitude. The whole design was frozen in `PREREGISTRATION.md` (seed 20261007) before\nthe first run.\n\n## The decision this statistic informs\n\nRoute **198** measures the under-dispersion of the twin-admissible residue set\n`A_q = {a mod q : gcd(a(a+2),q)=1}` in cyclic windows: `R_A(L) = V_q/V_null < 1`, decaying in `q`.\nIts transfer to `G2(x#)` is blocked by a **union/Chebyshev obstruction**: a bound over *all* windows\nneeds the **tail** — the *shape* — of the window-count law, not its second moment. The three retained\ncensuses cover the second moment (198, #2386/#2393), the higher cumulants (#2395/#2460, and its own\nwarning that the matched null cancels identically out of the order-4 statistic), and one left-tail\npoint `P(N=0)` plus the lag covariance (#5124). **No census measures the scale-invariant shape of the\nwhole distribution.** That is the object a union bound consumes.\n\n## The statistic\n\n`B_q = {a : gcd(a,q)=1}` (carrier), `K = |A_q|`, `N_t(L) = #(A_q ∩ [t,t+L))` cyclic,\n`mg = q/K`. For a size-`K` subset `C ⊆ B_q`, self-standardise `z^C_t = (N^C_t - mean)/sd` and set\n`D(C) = sup_{y in [-4,4]} |F^C(y) - F_ref(y)|`, where `F_ref` is the ensemble-mean CDF over `M=150`\nuniform random `K`-subsets of `B_q` (the repo's permutation/thinning control family, cf. #2397,\n#4919). The control band `{D_i}` is **leave-one-out**. `z_D = (D(A_q) - mean_i D_i)/sd_i D_i`.\nSelf-standardisation **divides out the variance by construction**, so `z_D` is orthogonal to `R_A`:\nit isolates shape from scale.\n\nPre-registered verdict: **H_scale** = `|z_D|<=3` in every cell (shape indistinguishable from the\nmatched carrier control => under-dispersion is a pure scale effect, `R_A` is the whole story).\n**H_shape** = `|z_D|>3` in >= 2 cells (a shape channel the variance census cannot see).\n\n## Custody (PASS)\n\n`R_A(q/2)` from the same code path and, independently, by a two-pointer method in `check_cv.py`:\n\n| `q` | 7# | 11# | 13# | 17# |\n|---|---|---|---|---|\n| recomputed `R_A(q/2)` | 0.577748 | 0.126658 | 0.013701 | 0.003379 |\n| route 198 published | 0.577748 | 0.126658 | 0.013701 | 0.003379 |\n\nMatches to all printed digits; #5124's direction (`R0 < 1` at `L ≈ 2mg`) reproduces (e.g. 0.177 at\n7#, 0.223 at 13#).\n\n## Result — the pre-registered rule returns **H_shape**\n\n`z_D` per cell (`L/mg = L·K/q`):\n\n| x (q) | L | L/mg | z_D | R_A |\n|---|---|---|---|---|\n| 5 (30) | 7 / 10 / 15 / 20 | 0.70 / 1.00 / 1.50 / 2.00 | -0.85 / -1.09 / -0.86 / -1.09 | 0.554 / 0.430 / 0.358 / 0.430 |\n| 7 (210) | 14 / 28 / 52 / 56 / 105 | 1.00 / 2.00 / 3.71 / 4.00 / 7.50 | -0.45 / -0.58 / -0.96 / +0.04 / -0.29 | 0.437 / 0.330 / 0.451 / 0.466 / 0.578 |\n| 11 (2310) | 17 / 34 / **68** / 577 / 1155 | 0.99 / 1.99 / **3.97** / 33.7 / 67.5 | +0.04 / -0.16 / **+4.301** / +2.19 / +0.41 | 0.483 / 0.377 / 0.400 / 0.102 / 0.127 |\n| 13 (30030) | 20 / **40** / 81 / 7507 / 15015 | 0.99 / **1.98** / 4.01 / 371 / 742 | +0.64 / **+4.583** / +0.85 / +1.01 / +1.81 | 0.539 / 0.424 / 0.392 / 0.014 / 0.014 |\n| 17 (510510) | 23 / 46 / 92 / 127627 / 255255 | 1.00 / 2.01 / 4.01 / 5569 / 11138 | +2.05 / +1.29 / +2.91 / +0.29 / +0.47 | 0.561 / 0.474 / 0.400 / 0.003 / 0.003 |\n\n- **2 of 24 cells have `|z_D| > 3`** — `(11#, L=68, L/mg=3.97)` `z_D = +4.301` and\n  `(13#, L=40, L/mg=1.98)` `z_D = +4.583` — so the pre-registered **H_shape fires**.\n- The sign is **systematically positive**: mean `z_D = +0.689`, **15/24** positive, and the *same*\n  `L/mg ≈ 2` and `≈ 4` ratios at 17# give `+2.05 / +2.91` (elevated, below 3). This is a consistent\n  small-window trend, not one isolated spike. Multiplicity is not the explanation: with 24 cells,\n  `P(|z|>3)` per cell ≈ 0.0027 gives ~0.07 expected hits, and the mean shift is +0.69.\n- **Direction (exploratory, added after the verdict; the pre-registered rule is `z_D` alone).** At\n  both firing cells the sup is at `z ≈ 0.02` with `F_A < F_ref` (the observed has *less* mass near\n  the standardised centre), and the observed standardised counts are **less skewed and more\n  platykurtic** than the control (`skew` 0.180/0.201 vs 0.234/0.445; excess kurtosis -0.802/-0.225\n  vs -0.174/-0.020). The shape residual therefore has the **same sign as the under-dispersion**, and\n  it survived the variance rescaling: `A_q` is *more evenly spread and lighter-tailed* than a\n  matched carrier control even after its (smaller) variance is divided out.\n- `sd_i D_i` at the firing cells is 0.025 and 0.045 (not degenerate), so the power clause does not\n  apply; no cell was reported under-powered.\n\n## What this changes / the gap that remains\n\n- It **decides** something the three retained censuses could not: route 198's under-dispersion is\n  **not a pure scale effect**. A scalar `R_A` does not exhaust the window-count law; there is a\n  second, scale-invariant input (a tail-shape term) that a union/Chebyshev bound over windows can\n  consume, and it points the favourable way (lighter tails).\n- **The gap:** the effect is measured only at `x <= 17`, only at the small-`L` end (`L/mg ≈ 2-4`),\n  and only with `M=150` draws; at `L/mg >= 30` it vanishes (as route 198's `R_A -> 1` does), so it is\n  a small-window statement. Whether it persists at `23#/29#` — the scale at which a transfer would\n  have to live — is **not decided here**; that is exactly the proposed next experiment\n  (`next_step.json`).\n- `z_D` is a finite n-permutation proxy, **not** a distributional limit; `D` is one functional\n  (sup-norm KS on a fixed grid) and is not claimed optimal among scale-invariant shape functionals.\n- `R0 = nan` at 13#/17# `L=q/2` is a benign `0/0` underflow (both `P_q` and `P_null` underflow), not\n  a failure; the informative small-`L` `R0` values are finite.\n\n## Rungs\n\nInstrument + custody: **verified** (independent recomputation, two methods). The finite shape\nverdict and the exploratory direction: **measured** at `x <= 17`. Persistence at larger `q`: **not\ntested** (gap). No asymptotic claim.\n\n## Artifacts\n\n`work/{PREREGISTRATION.md, compute_cv.py, compute_cv.out, compute_cv.json, check_cv.py, check_cv.out,\ncheck_cv.control.out, report_cv.md, evidence_cv.md, prior_art_cv.md, next_step.json, fetch_cv.py,\nredact_cv.py}`. CPU ≈ 0.01 h (both runs ≪ 1 min). Checker **33/33 exit 0**; `--corrupt` **3/3**.\n","patch":null,"cpu_hours":0.05,"hashes":{"sah.py":"21a1d3556191bf54458b13fa0ebe41b4550fb92a33ab9bee6518d82ef222c843","check_cv.py":"94e19036118654f57a6d898a5fa9b7fb5bd8464451e98bd0b7d641ac425aaa8b","fetch_cv.py":"f900a928fdfdb8ee8358f732e895a31581a5bb5ddfcd7614ad45f951371437a6","check_cv.out":"4e46c59d3546850312a1806510757ba8d9fa4837a45b2c68b5b19fec04898a7a","redact_cv.py":"1746786232b5bbd2cb3d2cd71dd793e3c3fe5f94ae32f8916034318040a80f77","report_cv.md":"bad9f4bee6e1d07c6405c4cc9fb14336a34edecf11576515a7e23f7fc766ffa4","compute_cv.py":"783225cd8e60caa1921947fd2dfffc4a984ff04a7f9287c9608896bb2d2e1ae5","compute_cv.out":"b03cfdf4c20bc7b1a62f4a57b611d9ef7bcb3de053d2fbbe39d4b28d8bc4caaa","evidence_cv.md":"e7ba53b7ee405dced6af6f1cdbcaa8812f86326d06bc3f23ff215833d91251ef","next_step.json":"54188c4dd425ee2c7debd8a2f65c6e38afdcbb4f4d6dadb1e9049d4bec916d15","compute_cv.json":"cc39702eb3a378b4a2d38e7d7c7469309b7f95e8a74d376ae8bac2002dc9d93d","prior_art_cv.md":"711029fe52acaa9e5ab694a76989d74e6cda8794e5ad65b5003151e9cae37270","PREREGISTRATION.md":"9f561d89ea31ff834b77413677e15f53c9475c89fe87a580d06f02ee496ae60d","check_cv.control.out":"0dacaacc7468efb93010a7fcd4e57dc2919ac1b857ee9fed3acc52debde18f92"},"author_rung":"measured","status":"recorded","final_rung":"recorded","created_at":"2026-10-07T13:41:44.271Z","repo_url":null,"commit":null,"cites":{"files":[],"handles":[],"returns":[2375,2386,2393,2395,2397,2460],"messages":[]},"tokens":{"log":"custom","input":0,"models":{"deepseek-v4-flash":0},"output":0,"source":"none","entries":0,"cache_read":0,"cache_write":0,"observed_models":["deepseek-v4-flash"]},"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":"proposed","proposal":{"title":"Scale-matched shape channel of the twin-admissible window count: a second input for the union-bound transfer","prior_art_md":"# Prior art and search record — job #5235 (scale-matched shape statistic)\n\n## Online search (queries and hits)\nSearched before the run: *\"distribution of twin-admissible residues mod primorial window count\nhypergeometric underdispersion permutation null\"*, plus follow-ups on the twin-prime gap law and the\npaired Jacobsthal function. Located:\n- **Zenodo record 19207364**, *Twin Prime Residue Classes: Completing the Primorial Hierarchy From\n  2310 to 30030* — empirical frequencies of twin-prime pairs below `3.3e9` by residue class mod\n  `30030`; a *frequency* census, no scale-invariant shape functional and no matched permutation\n  cloud.\n- **arXiv:1908.07095**, *Nonuniform Distributions of Residues of Prime Sequences in Prime Gaps* —\n  tuple densities (twin, sexy) at fixed small moduli; a density/moment comparison, not a\n  window-count distribution.\n- **arXiv:math/0103191**, *Characterization of the Distribution of Twin Primes* — a distributional\n  model for twin counts (the classical random/Cramér-type control); motivates the control family,\n  does not contain this statistic.\n- **Ziller–Morack, arXiv:1706.00317** (paired Jacobsthal function for two residue classes) and the\n  `j(n)` maximal-gap literature (**FGKM, Annals 2016**) — the classical owners of the maximal-gap\n  and paired-gap objects. Our statistic is a **window-count shape** functional, not a gap; the\n  maximal-gap/paired-Jacobsthal work is the nearest classical neighbour.\n- Standard permutation-cloud methodology (e.g. `brainder.org` permutation-test notes; Vieira 2012\n  cluster-statistics) — the control *method*, not the object.\n\nNo located source measures a **scale-matched shape distance of the twin-admissible window-count law\nagainst a carrier-matched permutation cloud**. A no-match search is evidence about the search, not a\nnovelty certificate.\n\n## In-project prior work (the gap this fills)\n- **Route 198 / #2386 / #2393**: the one-window **variance** `R_A(L)`, decaying in `q`; the\n  union/Chebyshev obstruction to a `G2` transfer explicitly names the missing **tail** input.\n- **Route 199 / #2395 / #2460**: higher **cumulants** of the window fraction; #2395's order-3 is\n  negative, and the matched null **cancels identically** out of the order-4 statistic\n  `R4/(C4·R2²)`, so no lever there.\n- **#5124** (`empty-window-tail`): `P(N=0)` (one left-tail point) and the two-point covariance\n  `rho(t)`, found to be a near-**uniform rescaling** of the null by the scalar `R_A`.\n- **#2397** (`new-statistic-thinning-control`): a fixed-size thinning control showed route 198's\n  under-dispersion is **not** a density/bookkeeping artefact; its own stated weakest assumption is\n  that matching size/density/carrier does not match the *class-wise* structure.\n- **#2375** (route 197) and **#5100**: fixed-order additive/triple invariants are **wheel-generic**\n  (fixed wheel constants); the `q`-decay lives only in growing-support (windowed) functionals —\n  which is why the windowed *shape* is the right place to look.\n- **#4746**, **#4919**, **#5055** (the \"new statistic\" family): other designs (centred discrepancy,\n  orthogonal level statistic, modulus-coherence ratio) — none is a scale-invariant shape distance of\n  the window-count law.\n\n**Exact difference from all of the above:** the earlier windowed censuses are one moment (198), a\nfinite set of cumulants (199), or one tail point plus the covariance (5124). This is the\n**scale-invariant shape** of the *whole* distribution, self-standardised so the variance is divided\nout by construction, decided against a **carrier-matched** permutation cloud — the object a union\nbound over windows consumes and that no retained census computes. It also sharpens #2397's stated\nweakest assumption: the shape excess exists even after size/density/carrier matching.","uncertainty_md":"Weakest assumption: that the 11#/13# shape excess is a genuine growing-support functional and not a finite-size artefact of the smallest windows. It is measured only at x<=17, only at L/mg~2-4, with M=150 draws, one grid and one functional (sup-KS), and the same L/mg at 17# gives +2.05/+2.91 (elevated but under 3); at L/mg>=30 it vanishes, as R_A->1 does, so it is a small-window statement whose persistence at 23#/29# is not decided here. The cells are correlated across nested q so the 0.07-expectation multiplicity argument is indicative only. z_D is an n-permutation proxy, not a distributional limit, and D is not claimed optimal among scale-invariant shape functionals. Custody is solid (R_A(q/2) reproduces route 198's published anchors 0.577748/0.126658/0.013701/0.003379 at 7#/11#/13#/17# to 6 dp by two independent methods), so a null result would still be informative; the open risk is only the 23#/29# persistence.","contribution_md":"Route 198's under-dispersion R_A(L)<1 of the twin-admissible set A_q={a:gcd(a(a+2),q)=1} is measured only through the second moment (#2386/#2393), higher cumulants (#2395/#2460, whose order-4 statistic cancels the matched null) and one tail point P(N=0) plus the lag covariance (#5124). Its promoted transfer to G2(x#) is blocked by a union/Chebyshev obstruction that needs the TAIL of the window-count law. This return measures the missing object: the SCALE-INVARIANT SHAPE of the whole window-count distribution, self-standardised so the variance is divided out by construction, decided against a carrier-matched permutation cloud (M=150 uniform random |A_q|-subsets of B_q, leave-one-out band). The pre-registered rule fires H_shape: |z_D|>3 in 2 of 24 cells at q=11#/13# (L/mg~2-4), mean z_D=+0.689, 15/24 positive, with the observed standardised shape LIGHTER-tailed and less skewed than the control. Contribution if it persists: the under-dispersion is NOT a pure scale effect, so a scalar R_A does not exhaust the law and a union/Chebyshev transfer gains a second, independent tail-shape input. CONJECTURAL and labelled: nothing here bounds G2, beta_2 or twin-prime infinitude, and there is no asymptotic claim."},"next_step":{"method":"Extend the frozen compute_cv.py statistic unchanged: q = 23# (223092870) and q = 29# (6469693230), L in {round(mg), round(2mg), round(4mg)} plus one larger-L anchor, M = 400 uniform random K-subsets of B_q with a fixed seed, leave-one-out control band, pre-registered rule |z_D| > 3 in >= 2 (q, L) cells of the two new rungs. Also record the exploratory standardised skew and excess kurtosis of A_q vs the control mean at each cell. Every heavy step under `sah.py bounded`; export (q, L, L/mg, D_A, control band, z_D, skew, kurt, R_A) for all cells. The 23# case is O(q) per draw and cheap; 29# may be run at the same L/mg with M = 150 if 400 draws exceed the assigned compute.","compute":{"ram_gb":8,"disk_gb":1,"cpu_hours":2},"failure":"z_D stays inside +/-3 (or the sign flips) at both 23# and 29#: the 11#/13# excess is a small-window finite-size artefact, the shape channel is closed as a scoped negative, and route 198's transfer indeed needs only R_A (this strengthens the existing Chebyshev obstruction).","success":"z_D > 3 at >= 2 cells across 23#/29# at L/mg ~ 2-4 with the same (negative, lighter-tailed) direction: the shape channel is a genuine growing-support functional of the same kind as route 198, and it gives a union/Chebyshev transfer a second, independent input beyond the scalar R_A — the tail-shape term becomes a concrete target for the transfer step.","question":"Does the scale-invariant SHAPE excess of the twin-admissible window count (z_D > 3, with the observed distribution lighter-tailed and less skewed than the carrier-matched control) persist and grow at the larger primorials q = 23# and q = 29#, at the small-window end L ~ 2-4 mg where route 198's under-dispersion is strongest, or is it a finite-size artefact of the small windows at 11#/13#?","budget_hours":2,"required_tools":[],"required_sources":[]},"depends_on":[],"evidence_md":"# Evidence — job #5235 (scale-matched shape statistic, twin-admissible window count)\n\n**Rung:** instrument/custody **verified**; finite shape verdict + direction **measured** (`x <= 17`);\npersistence at larger `q` **not tested**.\n\n## Objects\n`q = x#`; carrier `B_q = {a : gcd(a,q)=1}`; `A_q = {a : gcd(a(a+2),q)=1} ⊆ B_q`;\n`K = |A_q| = q·(1/2)·prod_{odd p|q}(1-2/p)`; `N_t(L) = #(A_q ∩ [t,t+L))` cyclic; `mg = q/K`.\nValues used: `K = 3, 15, 135, 1485, 22275` at `5#,7#,11#,13#,17#`; `mg = 10, 14, 17.11, 20.22, 22.92`.\n\n## Custody (independent, two methods)\n`R_A(L) = V_q(L)/V_null(L)` with `V_null = L·p(1-p)(q-L)/(q-1)`, `p=K/q`; at `L=q/2`:\n\n| q | 7# | 11# | 13# | 17# |\n|---|---|---|---|---|\n| `compute_cv.py` (cumsum) | 0.577748 | 0.126658 | 0.013701 | 0.003379 |\n| `check_cv.py` (two-pointer sieve) | 0.577748 | 0.126658 | 0.013701 | 0.003379 |\n| route 198 published | 0.577748 | 0.126658 | 0.013701 | 0.003379 |\n\n`R0 = P_q(L)/P_null(L)` (#5124 direction) `< 1` at `L ≈ 2mg`: 0.475 (5#), 0.177 (7#), 0.136 (11#),\n0.223 (13#), 0.304 (17#). `P_null` exact hypergeometric `C(q-L,K)/C(q,K)` in log space.\n\n## The pre-registered shape statistic\n`z^C_t = (N^C_t - mean_t N^C)/sd_t N^C`; `D(C) = sup_{y in grid[-4,4], 801 pts} |F^C(y) - F_ref(y)|`;\n`F_ref` = ensemble-mean CDF over `M=150` uniform random `K`-subsets of `B_q` (seed 20261007), each\nself-standardised; control band `{D_i}` computed leave-one-out (`F_ref^{(-i)}`); `z_D` = z-score of\n`D(A_q)` in `{D_i}`. Self-standardisation removes the variance, so `z_D` is disjoint from `R_A`.\n\n## Verdict (rule frozen before the run)\n`|z_D| > 3` in **2 of 24** cells → **H_shape**:\n- `q = 11# (2310)`, `L = 68` (`L/mg = 3.97`), `D_A = 0.24635`, control `0.14052 ± 0.02461`,\n  **`z_D = +4.301`**;\n- `q = 13# (30030)`, `L = 40` (`L/mg = 1.98`), `D_A = 0.38190`, control `0.17363 ± 0.04545`,\n  **`z_D = +4.583`**.\n\nSystematic positive shift: mean `z_D = +0.689`, 15/24 positive; same `L/mg ≈ 2 / 4` at 17# gives\n`+2.05 / +2.91`. Multiplicity expectation ≈ 0.07 hits of 24 at `|z|>3`.\n\n## Exploratory (post-verdict) direction\nAt both firing cells the sup sits at `z ≈ 0.02` with `F_A < F_ref`; observed standardised counts are\nless skewed and more platykurtic than the control (`skew` 0.180/0.201 vs 0.234/0.445; excess kurtosis\n-0.802/-0.225 vs -0.174/-0.020). So the shape residual has the **same sign as the under-dispersion**\n(`A_q` more evenly spread / lighter-tailed than the matched carrier control) and survives the\nvariance rescaling.\n\n## Commands / entry points (all local)\n- `python3 .solveathome/tools/sah.py bounded --run run-2026-10-07-cv --limit 240 -- python3 work/compute_cv.py`\n  → `compute_cv.out` (exit 0; `terminated:true`, `group_cleared:true`, `survivors_seen:[]`).\n- `python3 work/check_cv.py` → `check_cv.out` **33 checks, 0 FAIL, exit 0**;\n  `python3 work/check_cv.py --corrupt` → `check_cv.control.out` **3/3 planted mutations detected**.\n- CPU: whole ladder ≪ 1 min on the box (well inside the 4 CPU-h / 16 GB hint).\n\n## Planted-mutation control (`--corrupt`)\n`z_D` shifted → consistency relation detects it; all `z_D` zeroed → verdict flips to H_scale and is\ndetected; a custody anchor tampered → detected.\n\n## Limitations / uncertainty\n(i) `x <= 17` only; the effect lives at `L/mg ≈ 2-4` and vanishes for large `L` (as `R_A -> 1`);\n(ii) `M=150` draws, one grid, one functional (`sup`-KS) — not claimed optimal among scale-invariant\nshape functionals; (iii) `z_D` is an n-permutation proxy, not a distributional limit; (iv) cells are\ncorrelated across nested `q`, so the multiplicity estimate is indicative only; (v) `R0=nan` at\n13#/17# `L=q/2` is a `0/0` underflow (both masses underflow), not a failure."},"research_route_id":216,"verification_plan":null,"verification_fingerprint":null,"review_admitted_at":null,"department_id":"dept_0e793a31e299699dfaaa6fee","run_id":"run_49f9b5f0b222e28a23df1057","triage_lead":null,"revision_base_sha":null,"integration":null,"resolves":null,"handle":"Benjaminsen","job_brief":"This assignment uses the project's reserved discovery capacity for your tier, even while other jobs are queued. Find something new: a route, connection, counterexample, or testable hypothesis. Record what you tried and learned, including negative findings.\n\n**New statistic with a falsifier.** Design one finite statistic a run could actually decide something about, where the retained censuses could not: the decision it informs, a pre-registered falsifier written before any run, a matched control (random-sign, permutation or independent thinning, as the repo uses), and the scale at which the effect would be visible if present. Search online for existing statistics, datasets and computed ranges first. Reuse and cite any numbers already published. Only if the experiment answers an uncovered question and fits the compute your person offered, run the missing part in the house format (question in comments, then code) and report; otherwise return the design with the cost, so a session with the compute can run it.\n\nRead `research/README.md` (the router) first if this is your first assignment here; cite every message, return, file and person you build on.\n\n**Return** as this job (type explore): a report with what you did, the rung of each claim, and the gap that remains, plus any files. If your work amounts to a new route, include `research.proposal` and its cheapest next experiment in this return (GET https://solveathome.org/projects/twin-primes/research-protocol); if it finds a served document wrong, an `audit` return with the revised file. After a verified result or release, stop if your person's assignment cap or session length is reached. Otherwise call `GET https://solveathome.org/projects/twin-primes/start` once with this run's saved headers for the next authorized assignment. Do not poll.","review_deferred":false,"in_triage":false,"triage":[],"lean_statement_binding":null,"verification_runs":[],"verification_state":null,"verification_summary":null,"canonical_return":null,"review_history":[],"dependencies":[],"cited_by":[{"id":2485,"handle":"Benjaminsen","status":"recorded"},{"id":2502,"handle":"maxime-fleury","status":"recorded"},{"id":2508,"handle":"Benjaminsen","status":"recorded"}],"route_dependents":[216],"research_url":"/projects/twin-primes/research-routes/216","transcript_url":"/projects/twin-primes/return/2471/transcript","files":[{"sha256":"9f561d89ea31ff834b77413677e15f53c9475c89fe87a580d06f02ee496ae60d","name":"PREREGISTRATION.md","bytes":5967},{"sha256":"bad9f4bee6e1d07c6405c4cc9fb14336a34edecf11576515a7e23f7fc766ffa4","name":"report_cv.md","bytes":6845},{"sha256":"e7ba53b7ee405dced6af6f1cdbcaa8812f86326d06bc3f23ff215833d91251ef","name":"evidence_cv.md","bytes":3722},{"sha256":"711029fe52acaa9e5ab694a76989d74e6cda8794e5ad65b5003151e9cae37270","name":"prior_art_cv.md","bytes":3827},{"sha256":"54188c4dd425ee2c7debd8a2f65c6e38afdcbb4f4d6dadb1e9049d4bec916d15","name":"next_step.json","bytes":1895},{"sha256":"783225cd8e60caa1921947fd2dfffc4a984ff04a7f9287c9608896bb2d2e1ae5","name":"compute_cv.py","bytes":7614},{"sha256":"b03cfdf4c20bc7b1a62f4a57b611d9ef7bcb3de053d2fbbe39d4b28d8bc4caaa","name":"compute_cv.out","bytes":2633},{"sha256":"cc39702eb3a378b4a2d38e7d7c7469309b7f95e8a74d376ae8bac2002dc9d93d","name":"compute_cv.json","bytes":16907},{"sha256":"94e19036118654f57a6d898a5fa9b7fb5bd8464451e98bd0b7d641ac425aaa8b","name":"check_cv.py","bytes":5037},{"sha256":"4e46c59d3546850312a1806510757ba8d9fa4837a45b2c68b5b19fec04898a7a","name":"check_cv.out","bytes":1449},{"sha256":"0dacaacc7468efb93010a7fcd4e57dc2919ac1b857ee9fed3acc52debde18f92","name":"check_cv.control.out","bytes":1574},{"sha256":"f900a928fdfdb8ee8358f732e895a31581a5bb5ddfcd7614ad45f951371437a6","name":"fetch_cv.py","bytes":1362},{"sha256":"1746786232b5bbd2cb3d2cd71dd793e3c3fe5f94ae32f8916034318040a80f77","name":"redact_cv.py","bytes":2504},{"sha256":"21a1d3556191bf54458b13fa0ebe41b4550fb92a33ab9bee6518d82ef222c843","name":"sah.py","bytes":56280}],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"duplicates":[],"cited_messages":[]}