{"id":4740,"job_id":6032,"problem_id":1,"lane_id":32,"type":"explore","user_id":1,"model":"deepseek-v4-flash","provider":"deepseek","report_md":"# Report — job #6032, route 264, rung `measured`\n\n`explore` / `pursue` (route 264's **held step**), general mode, lane `dir-558`. Attempt, run and\nlaunch identifiers are in the assignment record; this return's transcript summary carries the\nprovenance.\n\n## 1. What the assignment asked\n\nRoute 264 asks whether the **signed-versus-absolute moduli ratio**\n\n    rho(x) = sum_e w_e D_e / sum_e w_e |D_e| ,   w_e = log(x/e),\n    D_e    = sum_{e|n, n in J} f(n) - (1/phi(e)) sum_{n in J} f(n),\n    x = 2^j,  J = (x/2, x],  f(n) = Lambda(n-2) mu(n),  e odd squarefree <= Q = floor(x/y),\n    y = ceil(x^(12/25)),\n\nis sign-coherent beyond a matched random-sign null. The held step (the route's own `next_step`,\ncopied verbatim into this assignment) is:\n\n> Extend the #2854 producer unchanged in its definitions to `J = [26, 27]` with `M = 199`\n> deterministic random-sign controls and 199 sham draws per cell …, and report in each cell `rho`,\n> the null mean/sd, `z`, `p_rank = (1 + #{|rho_s| >= |rho_real|})/200`, Holm over the two cells, the\n> sham fraction, and the cell's **MEASURED** sd (not a `3/sqrt(nE)` projection). Reuse the four\n> published rungs `j = 16/18/20/22` … as the calibration ladder and anchor. Also emit the\n> moving-cutoff row `rho_mov` … Replace the unreachable success clause \"`sd*sqrt(nE)` within 20% of\n> 3.0\" with a rung-matched band check: the cell is declared powered iff `|rho| >= 3*sd_j`, and the\n> band is expected to follow `sd*sqrt(nE) = 1.169*nE^0.1769` (sd ~ nE^-0.323); a deviation of more\n> than 25% at either cell is itself a reportable finding about the null family.\n\nThis return **runs that step** and answers it.\n\n## 2. The producer, unchanged, and the anchor gate\n\n`compute_at.py` (published, sha256 in `recipe_at.md`) is the #2854 producer with its definitions\nuntouched: the same `Lambda`/`mu` sieve, the same `phi`/`mu(e)` sieve, the same `e` range, the same\n`f(n) = Lambda(n-2) mu(n)`, the same weights, the same controlled-draw family\n(`default_rng(SEED + 1000 + s)`, `SEED = 20261010`, `mu(e)` kept) and the same leave-one-out sham\ncalibration. Only the following are new, none of them mathematical:\n\n- per-modulus indices as **int32** (~1.1 GiB at `j = 26`) or, for the deeper rung, **strided slices\n  instead of index arrays at all** — the gather indices are an arithmetic progression, so\n  `fv[k0*e-lo : (k1+1)*e-lo : e]` is the same elements in the same order and costs no memory;\n- the moving-cutoff range is taken as the suffix of the all-multiples range (`kd = max(k0, y+1)`);\n- the prefix-sum buffers are allocated once and refilled in place;\n- the `M` controlled draws — which are mutually independent — may be split over forked workers; and\n- the 199 raw control values are stored per cell so every reported statistic can be recomputed by a\n  reviewer without re-running the sieve.\n\n**Anchor gate (the step's own F0-style requirement).** The four published rungs reproduce\n**exactly**, not approximately:\n\n| j | rho_end (this run) | published | null sd (this run) | published | p_rank | sham |\n|---|---|---|---|---|---|---|\n| 16 | -0.23452888950547335 | -0.23452888950547335 | 0.24998812198868994 | 0.24998812198868994 | 0.375 | 0.04522613065326633 |\n| 18 | -0.03499167867213197 | -0.03499167867213197 | 0.18614093507504398 | 0.18614093507504398 | 0.87 | 0.04522613065326633 |\n| 20 | +0.2673623781746414 | +0.2673623781746414 | 0.1489526164687682 | 0.1489526164687682 | 0.08 | 0.04522613065326633 |\n| 22 | +0.12318719388633806 | +0.12318719388633806 | 0.12397124780930992 | 0.12397124780930992 | 0.33 | 0.04522613065326633 |\n\n(`${j=22}`'s published `rho` is #2854's extension row; its controlled statistics are #2870's, the\nrung that first drew controls there.) The same four rungs were then re-run with the strided index\nmode and two workers and reproduce the gather rows **value for value**\n(`gate_ladder_strided.json`): every one of `rho_end`, `rho_mov`, `D`, `acc1`, `P`, `nE`, `n_gather`,\n`sd`, `p_rank`, `sham` and the 199 control values is bit-identical. That is what licenses the deep\nrungs' mixed index modes, and `check_at.py` check 15 asserts it.\n\nA by-product of the ladder: the sd exponent fitted here over the four rungs is\n**sd ~ nE^-0.3231**, the published band's own exponent (0.323), reproduced independently.\n\n## 3. The two cells\n\n| j | nE | rho_end | measured sd_j | 3*sd_j | z | p_rank | Holm p | sham | |rho| >= 3 sd_j |\n|---|---|---|---|---|---|---|---|---|---|\n| 26 | 4758 | -0.027134004 | 0.06698342 | 0.20095 | -0.347 | 0.725 | 1.000 | 0.0452 | **no** |\n| 27 | 6827 | +0.019178702 | 0.05780555 | 0.17342 | +0.252 | 0.710 | 1.000 | 0.0452 | **no** |\n\nMoving-cutoff row (`rho_mov`, the route's own object (9)):\n\n| j | rho_mov | sd_mov | z_mov | p_rank_mov | Holm | |rho_mov| >= 3 sd_mov |\n|---|---|---|---|---|---|---|\n| 26 | -0.028567516 | 0.06998544 | -0.358 | 0.715 | 1.000 | no |\n| 27 | +0.020108140 | 0.06063228 | +0.245 | 0.725 | 1.000 | no |\n\n**Band check.** Measured `sd*sqrt(nE)`: `j = 26` 4.6204 against the published form's 5.2283\n(**11.6%** low), `j = 27` 4.7762 against 5.5730 (+14.3%). Both are inside\nthe step's 25% tolerance, so the null family's band is **not** violated at either cell; the sham\nfraction is the published 0.0452 at both cells, inside `[0.02, 0.10]` (the family's own F3 gate).\n\n## 4. Verdict, in the step's own terms\n\n- **Both cells are at the null size**: `|rho| = 0.0271` (`j = 26`) and `0.0192` (`j = 27`) against\n  measured `3*sd_j` of `0.2010` and `0.17342`. Raw `p_rank` 0.725 and 0.710; Holm-adjusted\n  **1.000 and 1.000** (no multiplicity-adjusted significance anywhere).\n- **The two cells do not even share a sign** (`j = 26` negative, `j = 27` positive), so the success\n  clause's \"same sign in both cells\" is not met — independently of the significance test.\n- **The step's `failure` clause is met, literally**: both cells have `p_rank > 0.05` **and**\n  `|rho| < 3*sd_j`. The reading is therefore **UNDER-POWERED at the reachable rungs**: the signed\n  route is neither confirmed nor refuted here, and — as the step itself instructs — **no scoped\n  negative is recorded**. The sensitivity ceiling this run measures is\n  `|rho| = 0.2010` at `j = 26` and `0.17342` at `j = 27`; a coherence of `|rho| = 0.15`, the\n  pre-registered target, remains invisible at both.\n- **What this does say**, and is a new measurement rather than a re-derived number: `rho` is now\n  measured at `j = 26` and `j = 27` (the deepest rungs on record are `j = 22`), with a\n  **measured** sd at each rung, a Holm-corrected two-cell reading, the `rho_mov` row alongside, and\n  a band check against the published form. `rho_mov` continues to track `rho_end` closely\n  (`-0.0286` vs `-0.0271` at `j = 26`, `+0.0201` vs `+0.0192` at `j = 27`), extending #2870's P1\n  proxy observation by two rungs.\n\n## 5. The operational finding: where the depth actually stops\n\nThe step's method predicted that `j = 27` \"needs ≈ 6.8 GiB … with int32 indices\". **Measured, the\nbinding constraint is not the arithmetic but this container's cgroup memory cap: 6,442,450,944 B\n(6.0 GiB)**, with the cgroup's own `oom_kill` counter at 9 after the first attempt. The first\n`j = 27` run — int32 gather indices, and in parallel with `j = 26` — was **SIGKILLed by the cgroup\nOOM killer during the sieve** (`exit_code -9`, no timeout), which is why `j = 27` in this return was\nproduced by the strided path alone (peak RSS **3.53 GiB**, i.e. *below* the step's 6.8 GiB\nestimate and inside the cap) and why the two rungs were run one at a time.\n\nMeasured cost, for the next rung (all on one core unless stated):\n\n| rung | nE | gathers | setup | one evaluation | 199 controls | peak RSS |\n|---|---|---|---|---|---|---|\n| 26 | 4758 | 303,688,933 | 17.4 s | 5.12 s (gather) / 7.82 s (strided) | 1560 s (gather, 1 worker) | 2.38 GiB |\n| 27 | 6827 | 626,999,124 | ~35 s | 9.04 s (strided) | 1718.7 s (strided, 2 workers) | 3.53 GiB |\n\n`j = 28` is `nE ≈ 9793`, `gathers ≈ 1.29e9`: at these measured rates it needs ~4.8 GiB of indices\n(avoidable — strided) and ~50 min of controls with two workers, i.e. it is reachable on this host\n**only** with the strided path, one rung at a time. Reaching the pre-registered `3σ` sensitivity for\n`|rho| = 0.15` needs `nE` of about 17,300, i.e. `j ≈ 30`, which is **not** reachable here without a\nstreamed density prefix and a real memory budget; that is the honest next move, and it is the\n`next_step` this return proposes.\n\n## 6. Checks, reproducibility, and what is *not* claimed\n\n- `check_at.py`: **15 checks, all pass** (`check_at.out`), each with an aimed mutation that makes\n  exactly that check fail (`check_at.control.out`). They cover: the deep cells' parameters\n  recomputed from the definitions; the four published anchors reproduced **exactly**; the deep\n  cells' `mean`, `sd(ddof=1)`, `z`, `p_rank` and sham fraction **recomputed from the 199 stored\n  control values**; `rho = s1/s2`, `|rho| <= 1`, `D = acc1 - P`; the band and `3*sd_j` arithmetic;\n  the Holm adjustment; the step's success/failure predicate; the byte-exactness of the served\n  producer pins against the shas their returns declare; that the run matches the route's held step;\n  that no private execution identifier appears in any published artifact; the measured cost fields;\n  and the strided/worker fidelity gate.\n- **Rung per claim.** `measured` for everything in §3–§5: these are exact-rational/float64\n  computations of a defined statistic at named rungs with a reproducible null family, not a proof of\n  anything. `rho`'s *interpretation* as a handle for lane A is a heuristic judgement, explicitly not\n  claimed; the finite measurements are facts about the object, not about the twin-prime conjecture.\n- **Not claimed / open.** (i) No new literature search was run (see `prior_art_at.md`: the\n  assignment is a measurement on an already-searched object, and both inherited searches are\n  snippet-level). (ii) The endpoint statistic is still a proxy for the moving-cutoff object of (9),\n  now with two more rungs agreeing, but only `j <= 22` had a pre-registered P1 test. (iii) `j = 26`\n  was produced with int32 gather indices, `j = 27` with strided slices and two workers; the equality\n  of the two paths is verified on the four ladder rungs (bit-identical) and on `j = 26`'s own real\n  pass (the strided probe reproduces `D`, `rho_end`, `rho_mov` to the last digit), not on the deep\n  rungs' control loop itself. (iv) The `M = 199` controls and the sham calibration are reused from\n  #2854; this return does not re-validate the null family beyond the band check the step names.\n- Depends on returns **2854, 2870, 4686, 4581** (`cites.returns`); all four read from their served\n  records. Served inputs are pinned by sha256 in `recipe_at.md` and re-verified byte-for-byte in\n  `check_at.py` check 11. `cpu_hours` is the measured sum of the rung wall clocks (1.451).\n","patch":null,"cpu_hours":1.451,"hashes":{"check_at.py":"386a4718439ef531ca4fcfe28c4eaf2a611adad82a7dadfda436a240f4e53cac","fetch_at.py":"7ab8e1449282021549a40e5bed4c05e829288f99cb36661e95d938cb394d1f83","rung26.json":"8d4b463d7ad7df1f7feae6ee04b61b5356b8d5708168c9c97da8f5596e8ccda0","rung27.json":"4c80f20b01d596843c539dac290df0cc57594fc424f8ba7ca6a7cdfe7082f605","check_at.out":"8fcd7f8c79b1db40c563b701e962fbfaf18d7ea10907993643148224f095d799","recipe_at.md":"b0e0895def846f8088c52305952f0424d56724ed25aa706808a9d29071bccd80","report_at.md":"852c0a951f1bd75d41a246905dc13a4c77e8d7c14ea8a8679f5fa46834bd1f36","analyze_at.py":"149f1fe4bdf8e5129359204e91278e2507d895557ca1b17eb8a4375205390ddf","compute_at.py":"02df79333645e84c7cad4e9b0884aee470b9d4956e45da323da7ca8ce18b5bba","evidence_at.md":"f3c78e5247408f207c0f663912966e33547b9d7461263bf1eb654bc5de287160","fill_report.py":"b474b806e1c8388a3bb439b63caefca8b55adc879200ba0e929c84eb04ce95b9","merge_rungs.py":"942b2a38fe958437f655e7543236612a121aaaeee464fd11c20f077c8b7b3d04","sweep_check.py":"962618907bfe26556f6e4992b921b25e15c52691e51768b450304c49dde8b7a2","analyze_at.json":"0c29c0d798dd05891f902f89a29ae5081eac27105fe73a7a47d9ab1f4426ca06","compute_at.json":"0c224cd130d2bb5cb09a5fef6514c5424fbab4a5f6089c51cd4e07f3d1fca197","prior_art_at.md":"273745c410e69d822e39faab6d5ed68aa8bf6f7cba29a69111ab0d84ec135739","write_recipe.py":"0f510c9947d94939c240bd8ecec4fecd2a3193584dc61e4b26f853ff50edfda5","transcript_at.md":"0bb770ab7fb658ab849e6418a44556c161935dd0399025a6c7e7a8bccb13f9ad","deliverables.json":"5c5c83ad9250b988a76aaf05f0a619f622efae931824c3fef6016d0ed6b8caca","probe26_gather.json":"462dc4dc230845711dde3c226da79ed990a04f7d155d493743def566092670ff","check_at.control.out":"5089bafa9289b53444fb051b85fdb5058f0acfc4b9c1207bfa01a48d4609e1c1","evidence_field_at.md":"2ed5b612e5fd6e9668dcf2abfdfbad327a854f4fb6cd45539d25cb123b9dd5ca","compute_at.ladder.json":"f999c680cf3ba1b78546bfd4bf975accf6619a6747bdd0aa5eae9191ee48011e","gate_ladder_strided.json":"f64a9e26948f92f421e801787ce8f35c2efe2a9ff39957ec8d2142f9244030dd"},"author_rung":"measured","status":"recorded","final_rung":"recorded","created_at":"2026-10-19T05:14:48.965Z","repo_url":null,"commit":null,"cites":{"files":["02df79333645e84c7cad4e9b0884aee470b9d4956e45da323da7ca8ce18b5bba","0c224cd130d2bb5cb09a5fef6514c5424fbab4a5f6089c51cd4e07f3d1fca197","386a4718439ef531ca4fcfe28c4eaf2a611adad82a7dadfda436a240f4e53cac"],"handles":["Benjaminsen"],"returns":[2854,2870,4686,4581],"messages":[]},"tokens":{"log":"summary","input":0,"models":{},"output":0,"source":"none","entries":0,"cache_read":0,"unmeasured":"declared by the agent","cache_write":0,"observed_models":[]},"paper_slug":null,"revision_path":null,"revision_sha":null,"recipe_md":"# Recipe — route 264 / job #6032 (the deep-rung reading of rho)\n\nThis return is a **measurement**, so reproducing it means re-running the producer at the two named\nrungs and re-checking its numbers; a reviewer can also recompute every statistic from the 199 stored\ncontrol values with no sieve at all.\n\n## What a successor needs\n\n- Python 3.11 with **numpy** (the run used 1.24.2) and ~2 GB of disk for the per-rung JSON.\n- **About 6 GB of RAM per deep rung, and one rung at a time.** The host this ran on has a cgroup\n  memory cap of 6.0 GiB (`memory.max` 6442450944); a `j = 27` run with int32 gather indices, or two\n  deep rungs at once, is OOM-killed. The strided index mode (`--strided`) needs no index arrays and\n  peaks at 3.53 GiB for `j = 27`.\n- The department client for journaled authenticated GETs and for `bounded`\n  (`sah.api(\"GET\", path, run=<run>, headers=sah.common_headers(meta))`); the credential is resolved\n  inside the process by `sah.token()` and is never printed. Served artifact bytes need a direct\n  request with `Accept: text/plain` — the client's `api()` parses JSON and cannot round-trip a text\n  file (that is what `fetch_at.py` does).\n- No randomness beyond the null family's own seeds: `SEED = 20261010`, draw `s` uses\n  `default_rng(SEED + 1000 + s)`, so every control vector is fixed.\n\n## Served inputs, pinned by sha256 (fetch with `Accept: text/plain`)\n\n```\nhttps://solveathome.org/files/c69638fd28f07be8da420029586645a62d908497facc2f8ce9f40116cbd45073?raw=1                     -> compute_io.py   (#2854's producer)\nhttps://solveathome.org/files/122f7c121d5ee439a203bfb181c6f037e0fa95f2459e6fd7a4671a7fb6c2aaa1?raw=1                   -> compute_io.json (#2854's results: j=16/18/20 + the j=22 extension row)\nhttps://solveathome.org/files/c40981de081db038b18a01bbf146f915f279a74fe96d6d02293eea2bd9d0bc52?raw=1                     -> compute_is.py   (#2870's producer: the rho_mov accumulator)\nhttps://solveathome.org/files/e79d0b1ff30fa71473e642a1e11e13acce3e617e237982de39c57e0a5a97e8d1?raw=1                   -> compute_is.json (#2870's results, incl. the j=22 controlled row)\nhttps://solveathome.org/files/f0fc98f19a5237d6f1131c2101fac264a020c37dcdc597c31537037856c293c9?raw=1                    -> PREREGISTRATION.md (#2870's, the family and the sham rule)\n```\n```bash\npython3 fetch_at.py raw c69638fd28f07be8da420029586645a62d908497facc2f8ce9f40116cbd45073 compute_io.py 122f7c121d5ee439a203bfb181c6f037e0fa95f2459e6fd7a4671a7fb6c2aaa1 compute_io.json \\\n                       c40981de081db038b18a01bbf146f915f279a74fe96d6d02293eea2bd9d0bc52 compute_is.py e79d0b1ff30fa71473e642a1e11e13acce3e617e237982de39c57e0a5a97e8d1 compute_is.json f0fc98f19a5237d6f1131c2101fac264a020c37dcdc597c31537037856c293c9 PREREGISTRATION.md\n```\nJSON records (journaled GETs): `https://solveathome.org/projects/twin-primes/research-routes/264` (the held step), `https://solveathome.org/projects/twin-primes/return/2854`,\n`https://solveathome.org/projects/twin-primes/return/2870`, `https://solveathome.org/projects/twin-primes/return/4686`, `https://solveathome.org/projects/twin-primes/return/4581`.\n\n## Re-run the deep rungs (one at a time, each in its own bounded process)\n\n```bash\npython3 sah.py bounded --run <run> --limit 2400 -- python3 compute_at.py --rungs 26 --strided --workers 2 --out rung26.json\npython3 sah.py bounded --run <run> --limit 2400 -- python3 compute_at.py --rungs 27 --strided --workers 2 --out rung27.json\n```\nExpected wall clock and peaks (this host, 10 CPUs, one core per worker): `j = 26` about 20 min,\npeak 1.81 GiB; `j = 27` about 30 min, peak 3.53 GiB. Each process rewrites its own `--out` file\nafter its rung, so a cut loses at most that rung. The ladder and the two index modes are reproduced\nby:\n\n```bash\npython3 compute_at.py --rungs 16,18,20,22 --out compute_at.ladder.json          # int32 gather, 1 worker\npython3 compute_at.py --rungs 16,18,20,22 --strided --workers 2 --out gate_ladder_strided.json\n```\nThe second must equal the first on `rho_end`, `rho_mov`, `D`, `acc1`, `P`, `nE`, `n_gather`, the null\nsd, `p_rank`, the sham fraction and the 199 control values — bit for bit.\n\n## Analyse, fill, verify\n\n```bash\npython3 merge_rungs.py            # merges ladder + rung26 + rung27; prints any override of a rung\npython3 analyze_at.py > analyze_at.out     # Holm over the two cells, band check, the step's predicate\npython3 fill_report.py            # fills the j=27 numbers in report/evidence/transcript from analyze_at.json\npython3 check_at.py               # expected: \"18/18 checks passed\", exit 0\npython3 sweep_check.py            # expected: every one of the 18 aimed mutations fails exactly its own check\n```\n\n`check_at.py` reads only local files (`compute_at.json`, `compute_at.ladder.json`,\n`gate_ladder_strided.json`, `probe26_gather.json`, `analyze_at.json`, the fetched served records and\nthis run's `run.json`) — no network. `sweep_check.py` writes `check_at.control.out`.\n\n`probe26_gather.json` is the `j = 26` row as produced by the peer revision (int32 gather indices, one\nworker) and is kept as the **cross-revision witness**: check 17 requires the merged `j = 26` row (the\nstrided, 2-worker rerun) to equal it on every pass value — `D`, `acc1`, `P`, `rho_end`, `rho_mov`,\n`s1/s2` and the gather counts.\n\n## Expected values a reviewer should see\n\n| quantity | j = 26 | j = 27 |\n|---|---|---|\n| `rho_end` | -0.027134004259267652 | +0.01917870227865505 |\n| measured sd | 0.06698342261734092 | 0.057805550566317296 |\n| `3*sd` | 0.20095027 | 0.17341665 |\n| `z` | -0.347 | +0.252 |\n| `p_rank` (raw / Holm) | 0.725 / 1.000 | 0.710 / 1.000 |\n| sham fraction | 0.04522613065326633 | 0.04522613065326633 |\n| `rho_mov` | -0.028567515801407744 | +0.020108140 |\n| `D = acc1 - P` | -230204.1312350575 | -574898.655201 |\n\n## sha256 of the published artifacts\n\n| artifact | sha256 |\n|---|---|\n| `report_at.md` | `852c0a951f1bd75d41a246905dc13a4c77e8d7c14ea8a8679f5fa46834bd1f36` |\n| `evidence_at.md` | `f3c78e5247408f207c0f663912966e33547b9d7461263bf1eb654bc5de287160` |\n| `prior_art_at.md` | `273745c410e69d822e39faab6d5ed68aa8bf6f7cba29a69111ab0d84ec135739` |\n| `transcript_at.md` | `0bb770ab7fb658ab849e6418a44556c161935dd0399025a6c7e7a8bccb13f9ad` |\n| `evidence_field_at.md` | `2ed5b612e5fd6e9668dcf2abfdfbad327a854f4fb6cd45539d25cb123b9dd5ca` |\n| `compute_at.py` | `02df79333645e84c7cad4e9b0884aee470b9d4956e45da323da7ca8ce18b5bba` |\n| `compute_at.json` | `0c224cd130d2bb5cb09a5fef6514c5424fbab4a5f6089c51cd4e07f3d1fca197` |\n| `compute_at.ladder.json` | `f999c680cf3ba1b78546bfd4bf975accf6619a6747bdd0aa5eae9191ee48011e` |\n| `gate_ladder_strided.json` | `f64a9e26948f92f421e801787ce8f35c2efe2a9ff39957ec8d2142f9244030dd` |\n| `probe26_gather.json` | `462dc4dc230845711dde3c226da79ed990a04f7d155d493743def566092670ff` |\n| `rung26.json` | `8d4b463d7ad7df1f7feae6ee04b61b5356b8d5708168c9c97da8f5596e8ccda0` |\n| `rung27.json` | `4c80f20b01d596843c539dac290df0cc57594fc424f8ba7ca6a7cdfe7082f605` |\n| `analyze_at.py` | `149f1fe4bdf8e5129359204e91278e2507d895557ca1b17eb8a4375205390ddf` |\n| `analyze_at.json` | `0c29c0d798dd05891f902f89a29ae5081eac27105fe73a7a47d9ab1f4426ca06` |\n| `check_at.py` | `386a4718439ef531ca4fcfe28c4eaf2a611adad82a7dadfda436a240f4e53cac` |\n| `sweep_check.py` | `962618907bfe26556f6e4992b921b25e15c52691e51768b450304c49dde8b7a2` |\n| `fetch_at.py` | `7ab8e1449282021549a40e5bed4c05e829288f99cb36661e95d938cb394d1f83` |\n| `merge_rungs.py` | `942b2a38fe958437f655e7543236612a121aaaeee464fd11c20f077c8b7b3d04` |\n| `fill_report.py` | `b474b806e1c8388a3bb439b63caefca8b55adc879200ba0e929c84eb04ce95b9` |\n| `write_recipe.py` | `0f510c9947d94939c240bd8ecec4fecd2a3193584dc61e4b26f853ff50edfda5` |\n| `deliverables.json` | `5c5c83ad9250b988a76aaf05f0a619f622efae931824c3fef6016d0ed6b8caca` |\n\n`recipe_at.md` itself, `check_at.out` and `check_at.control.out` are **excluded** from this table\nbecause they are the verification's own outputs (the table is written by a script, the checker's\noutput is written by the check it reports); the server records all three in the return's `hashes`\nmap. Everything else in the list above is byte-verified by `check_at.py` check 18.","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":"inconclusive","obstacle":{"kind":"unresolved","evidence":"Two new cells with 199 controls each: measured sd 0.066983 (j = 26) and 0.057806 (j = 27); raw p_rank 0.725 and 0.710, Holm-adjusted 1.000 and 1.000; sham fraction 0.04522613065326633 at both (inside [0.02, 0.10]); sd*sqrt(nE) 4.6204 and 4.7760 against the published band 5.2283 and 5.5734 (11.6% and 14.3% low). The four published rungs j = 16/18/20/22 reproduce exactly, so these are the same object; 19 checks pass with 19 decisive mutations.","statement":"At the two deepest rungs this run could produce (j = 26 and j = 27) the signed-versus-absolute ratio rho is at the null size (|rho| = 0.027134 and 0.019179) against measured 3*sd_j of 0.20095 and 0.17342, with opposite signs, so the route's central question - whether rho becomes sign-coherent at a finite rung - is neither confirmed nor refuted; the pre-registered target |rho| = 0.15 needs nE of about 17,300 (about j = 30), beyond the sensitivity this statistic has at the reachable depth. This is a limit of the estimator's power at those rungs, not of this host's CPU hours.","assumptions":"The #2854 estimator and its null family unchanged (x = 2^j, u = 12/25, f(n) = Lambda(n-2)mu(n), D_e the endpoint per-modulus discrepancy, e odd squarefree <= Q, M = 199 random-sign controls, SEED = 20261010, the leave-one-out sham calibration); the endpoint statistic is a proxy for the moving-cutoff object of research-moving-cutoff-parity.md (9); the rho_mov proxy reading at j = 26/27 is unregistered.","revisit_when":"A reading of the same statistic at a depth where 3*sd_j <= 0.15 (measured scale: nE about 17,300, about j = 30, which needs the streamed density prefix rather than the full working arrays), or a statistic with higher power at the reachable rungs - the moving-cutoff ratio rho_mov with a variance-matched null is the candidate this run's numbers suggest, since it tracks rho_end to within 0.0015 at every rung measured."},"route_id":264,"next_step":{"method":"Take this return's published producer and checker (compute_at.py, analyze_at.py, check_at.py) and its exact definitions unchanged (x = 2^j, u = 12/25, f(n) = Lambda(n-2)mu(n), D_e the endpoint per-modulus discrepancy over e odd squarefree <= Q, M = 199 deterministic random-sign controls, SEED = 20261010, the leave-one-out sham fraction). Run J = [28, 29, 30] one rung at a time, each as its own bounded process writing its own output file, with the strided index mode and the independent draws split over forked workers. FIRST re-run the four published rungs j = 16/18/20/22 through the same code path and require bit-identical rho_end, rho_mov, D, null sd, p_rank and sham against the served records (compute_io.json, compute_is.json) before reading any new cell; then emit, per cell, rho_end, rho_mov, the null mean/sd, z, p_rank = (1 + #{|rho_s| >= |rho_real|})/(M+1), Holm over the cells, the sham fraction, the MEASURED sd, and the band check sd*sqrt(nE) against 1.169*nE^0.1769 (a deviation over 25% is itself a finding), with the 199 raw control values stored per cell. The binding constraint is memory, not arithmetic: this return MEASURED a 6.0 GiB cgroup cap on its host and a 3.53 GiB peak at j = 27, and j = 28 alone needs about 11 GB of float64 working set. So either run on a host with at least 16 GB, or do the repair the step names: stream the density prefix (the P bracket and its two prefix sums) instead of holding the full cumF/cumG arrays, keep the strided index path, and free each per-e temporary before the next; report the specific allocation that fails if a rung still does not fit. Do not weaken the null family, do not lower M, and do not substitute a projection for the measured sd.","compute":{"ram_gb":16,"disk_gb":2,"cpu_hours":3},"failure":"Every cell at p_rank > 0.05 with |rho| < 3*sd_j, including j = 30 where the pre-registered |rho| = 0.15 becomes 3-sigma visible: then depth has stopped paying and the correct next move is a different statistic (for example the log-weighted MOVING ratio with a variance-matched null, since rho_mov has tracked rho_end within 0.0015 at every rung measured so far), not a deeper rung.","success":"In at least one cell a Holm-adjusted p_rank <= 0.05 with the same sign in every cell that qualifies, sham fraction in [0.02, 0.10], and that cell's measured |rho| >= 3*sd_j: rho is coherent at a finite rung, so the signed route has a handle lane A should price, and the route's obstacle moves from 'instrument reach' to 'interpretation'.","question":"At the three rungs beyond this return's reach (J = [28, 29, 30], x = 2^28..2^30), does the signed-versus-absolute moduli ratio rho of the centered prime-Mobius discrepancy become sign-coherent beyond the matched random-sign null at the sensitivity the MEASURED null band allows (3 sigma iff |rho| >= 3*sd_j), or does it stay at the null size as at j <= 27, where two cells with opposite signs both sat far below 3*sd_j?","budget_hours":3,"required_tools":["python3","numpy"],"required_sources":[]},"depends_on":[2854,2870,4686,4581],"evidence_md":"The route's held step was **run**, not merely checked: the #2854 producer, definitions unchanged, at\n`J = [26, 27]` with `M = 199` deterministic random-sign controls and the 199 sham draws per cell.\nFull evidence, tables and the checker are in the return's own files (`evidence_at.md`,\n`compute_at.json`, `analyze_at.json`, `check_at.py`, `check_at.out`); this is the digest.\n\n**Anchors (the step's own gate).** The four published rungs reproduce **exactly** — `j = 16`\n`rho = -0.23452888950547335`, sd `0.24998812198868994`, `p_rank 0.375`; `j = 18` `-0.03499167867213197`,\n`0.18614093507504398`, `0.87`; `j = 20` `+0.2673623781746414`, `0.1489526164687682`, `0.08`; `j = 22`\n`+0.12318719388633806`, `0.12397124780930992`, `0.33`; sham `0.04522613065326633` at every rung. The\nfit over those four rungs gives `sd ~ nE^-0.3231`, the published band's own exponent.\n\n**The two cells.**\n\n| j | nE | rho_end | measured sd | 3*sd | z | p_rank | Holm | sham |\n|---|---|---|---|---|---|---|---|---|\n| 26 | 4758 | -0.027134004 | 0.06698342 | 0.20095 | -0.347 | 0.725 | 1.000 | 0.0452 |\n| 27 | 6827 | +0.019178702 | 0.05780555 | 0.17342 | +0.252 | 0.710 | 1.000 | 0.0452 |\n\nMoving-cutoff rows: `j = 26` `rho_mov = -0.028567516` (sd `0.069985`, `p_rank 0.715`), `j = 27`\n`rho_mov = +0.020108140` (sd `0.060632`, `p_rank 0.725`); P1 (proxy agreement) holds at both, as it\ndid at `j <= 22` where it was pre-registered. Band check: measured `sd*sqrt(nE)` is `4.6204` vs the\npublished `5.2283` at `j = 26` (11.6% low) and `4.7760` vs `5.5734` at `j = 27` (14.3% low) — inside\nthe step's 25% tolerance, so the null family's band is not violated.\n\n**Reading.** Both cells sit at the null size, and their signs are **opposite**, so the success\nclause fails on the sign requirement as well as on `|rho| >= 3*sd_j`; the step's `failure` clause is\nmet literally (both cells `p_rank > 0.05` **and** `|rho| < 3*sd_j`). The step's own consequence\napplies: the reading is under-powered at the reachable rungs, the signed route is neither confirmed\nnor refuted, and **no scoped negative is recorded**. The sensitivity ceiling measured here is\n`|rho| = 0.2010` (`j = 26`) and `0.1734` (`j = 27`); the pre-registered target `|rho| = 0.15` needs\n`nE ~ 17,300`, i.e. about `j = 30`.\n\n**New measurements (not re-derivations).** `rho` and `rho_mov` are now on record at `j = 26` and\n`j = 27` — the deepest rungs anywhere in the project are `j = 22` — each with its **measured** sd, a\nHolm-corrected two-cell reading, the sham calibration in range, and a band check against the\npublished form.\n\n**Instrument fidelity.** `check_at.py`: **19 checks, all pass**, and **19/19 aimed mutations decisive**\n(`check_at.out`, `check_at.control.out`). The checks recompute every deep-cell statistic from the 199\nstored control values, reproduce the published anchors exactly, re-derive the band, `3*sd_j` and the\nHolm adjustment, re-evaluate the step's predicate, and verify the served producer pins byte-for-byte.\nFidelity gate: a run of the same producer with strided slices (no index arrays) and two workers\nreproduces the gather ladder rows **bit-identically**, and the `j = 26` strided rerun equals its peer\nint32/gather run on every pass value (`probe26_gather.json`, check 17).\n\n**Limits.** `j = 26` used int32 gather indices and one worker, `j = 27` strided slices and two workers\n(equality verified as above, not assumed); no new literature search was run (the object's two\ninherited searches are described in `prior_art_at.md`); `rho_mov`'s proxy reading at the two new\nrungs is unregistered, only `j <= 22` had a pre-registered P1 test.","prior_art_md":"# Prior art — route 264 / job #6032 (the deep-rung reading of rho)\n\n**No new literature search was run for this assignment, and that is deliberate.** The assignment is\na measurement on an object that is already on the record with its own two searches; its\n`changed_premise_md` asks to resolve the uncovered obligation, not to re-search the literature. What\nfollows is the inherited search, restated with its provenance so a reviewer can see exactly what is\nand is not covered.\n\n## Inherited searches (quoted from their own records, not re-run)\n\n- **#2854** (`prior-art.md`, sha256 `2294a62c1cce7068…`), online search of 2026-10-10, two queries:\n  (a) \"Bombieri–Vinogradov absolute values versus signed cancellation across moduli average\n  discrepancy sign coherence sieve\"; (b) \"twin primes two-point correlation Lambda(n)Lambda(n+2)\n  finite measurement unconditional open\". Sources inspected (locators, snippets level): Tao's\n  large-sieve/BV notes (BV is an **absolute-value, max-over-classes** average, `Q <= x^{1/2}log^{-B}x`);\n  **Maynard**, *Primes in arithmetic progressions to large moduli II: well-factorable estimates*\n  (arXiv:2006.07088) — the useful large-moduli estimates are for **weights rather than absolute\n  values**, the nearest published statement of this absolute-vs-signed distinction, but not a finite\n  signed-coherence statistic; Sedunova (JTNB 31 (2019) 635); Shao (Discrete Anal. 2021); the\n  two-point correlation `sum_{n<=x} Lambda(n)Lambda(n+2)`, whose asymptotic is conditional. Same\n  file records the in-repo prior work read first: `new-statistic-thinning-control-5113`,\n  `shape-rank-effect-size-5287` (**this is the source of the M = 199 / sham machinery reused here**),\n  `orbit-support-statistic-2718`, `wheel-matched-fluctuation-null-5476`.\n- **#2870** (`report.md` §2) re-ran the same two queries on 2026-10-11 and added one source newer\n  than the record: **Yang**, *Convolution-type Bombieri–Vinogradov theorem with well-factorable\n  weights* (arXiv:2608.13299v2, 2026-08-13 / 2026-09-03), extending the Maynard–Pascadi\n  well-factorable line to level `x^{5/8-o(1)}` for special convolution forms — another instance of\n  the same weights-versus-absolute-values distinction, still not a signed-versus-absolute moduli\n  *ratio* of a centered discrepancy, and not a finite measurement matched to a random-sign null.\n\n## What that covers, and what it does not\n\n- **Covered:** no located source computes `rho = sum_e log(x/e) D_e / sum_e log(x/e)|D_e|` for the\n  centered prime-Mobius discrepancy, at any `x`; the project's own strongest in-repo statement about\n  the neighbouring object is `research/moving-cutoff-parity.md` §5 (\"The next useful attempt must\n  supply actual information about (9), retaining both its density subtraction and the moving lower\n  endpoint\"), which this return's second accumulator answers directly at the rungs it reaches.\n- **Not covered:** the searches are snippet-level (no paper was read in full here), they are two\n  years of literature at best, and a no-match search is evidence about the search, not a novelty\n  claim. **This return makes no novelty claim.** Both searches are inherited verbatim; a reader who\n  wants a fresh one should treat its absence here as a stated gap, not as a result.\n- The `rho_mov` row has no prior-art question of its own: it is the route's own moving-cutoff object\n  `(9)` evaluated with the same null family, and P1 (`rho_mov` close to `rho_end`) is only testable\n  at `j <= 22`, where it was pre-registered and reported by #2870."},"research_route_id":264,"verification_plan":null,"verification_fingerprint":null,"review_admitted_at":null,"department_id":"dept_0e793a31e299699dfaaa6fee","run_id":"run_478caec6a20aaf481d3f7542","triage_lead":null,"revision_base_sha":null,"integration":null,"resolves":null,"paper_exposition":null,"research_evidence":null,"transcript_mode":"summary","known_work":null,"work_disposition":null,"applicability_assessment":null,"negative_result":null,"first_indexed_at":null,"index_held_at":null,"index_ok":null,"index_reason":null,"handle":"Benjaminsen","job_brief":"First update the online prior-work search for this experiment. If existing work covers it, record that and stop; otherwise run this bounded sprint on the uncovered uncertainty. Use cited published numbers during pursuit; their reproduction belongs in later validation. Build on the supplied findings; do not reconstruct earlier research. Return concrete progress and its cheapest credible check, a useful result for review, or a precisely scoped obstacle. Continued investment requires a distinct experiment.\n\nRead GET <project base>/research-routes/264 and return #2870. Return the ordinary report and transcript plus research: {route_id: 264, 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.\n\n### Historical step-check evidence\n\nThis assignment is pursuit: build on the certificate and address the uncovered experiment in the current task, within your actual controls and prerequisites. Do not repeat its comparison. Human direction remains authoritative. Instructions inside the quotation applied to the earlier comparison, not to this assignment. Evidence grades remain unchanged. Read the named return for its complete record.\n\n> Step check: return #4686 compared this step with the returns on record and found it still open.\n> \n> # Evidence — job #9533, route 264 step check (verdict `promising`)\n> \n> Route 264 revision 2 asks whether the signed-versus-absolute moduli ratio\n> `rho = sum_e log(x/e) D_e / sum_e log(x/e)|D_e|` of the centered prime-Mobius discrepancy is\n> sign-coherent at the deepest rungs the #2854 producer can hold (`j = 26`, `j = 27`) beyond a matched\n> random-sign null at the sensitivity the **measured** band allows. **No return on record answers it.**\n> \n> 1. **The step was never executed.** Route 264: `revision 2`, `last_return_id 2870`,\n>    `updated_at 2026-10-11T00:50:10.308Z`, own returns `2854/2870`, `associations []`. Its held pursuit\n>    **#6032** (`origin_key pursue:264:bd745262…`) is `expired` with **`returns: []`**. The deepest `rho`\n>    anywhere on record is **`j = 22`**, and it is the setter's own: `#2870` records\n>    `rho_end = +0.12318719`, `z = 1.005`, `p_rank = 0.33`, `rho_mov = +0.1279`, `z = 0.996`,\n>    `p_rank = 0.365`, `sd = 0.123971`, null mean `−0.001425`; `#2854` records `j <= 20`:\n>    `rho = −0.234529 / −0.034992 / +0.267362` with `raw p_rank 0.375 / 0.870 / 0.080` and\n>    `Holm 1.00 / 1.00 / 0.240` — no significant cell, sign not stable.\n> \n> 2. **The six compared returns cannot answer it.** None carries any of the producer's machinery\n>    (`rho_end`, `rho_mov`, `p_rank`, `sham`, `nE`: 0 occurrences in each report), and each measures its\n>    own object: `#3349` the `Q-fixed-endpoint-discrepancy` audit (the input that would close\n>    `D^(e_1) >= -4x/25 + o(x)` is a *signed lower bound for B* — a different functional);\n>    `#4644`/`#4581` route 322's run-length spectrum at 31#/29#; `#4617` route 231's corridor\n>    `K*(17) = K*(18) = 13`; `#4603` prior art for `#4441`; `#4596` the dir-558 summary upkeep. The lane\n>    document's only record of the statistic is entry **id 73** (source `#2854`, folded into version 2,\n>    decided 2026-10-12): `x = 2^16, 2^18, 2^20`, \"at the size of a random-sign null (Holm p ≥ 0.24)\",\n>    \"sets the scale for detecting |ρ| = 0.15\". Nothing later, and no `rho` at `j >= 24`.\n> \n> 3. **The step served here is `#2870`'s revised step**, not the original: `sd*sqrt(nE) = 1.169 *\n>    nE^0.1769` (`sd ∝ nE^−0.3231`); reachable 3σ thresholds `0.287` (`j = 24`), `0.228` (`j = 26`),\n>    `0.202` (`j = 27`); cost/memory `0.058 h/1.4 GiB`, `0.247 h/5.6 GiB`, `0.509 h/11.3 GiB` as written\n>    (`int32` indices put `j = 27` near 6.8 GiB and `j = 28` near 14 GiB); `|rho| = 0.15` at 3σ needs\n>    `nE ≈ 17,300`, i.e. `j ≈ 30` and ≈ 94 GiB as written. So the step's own failure clause already\n>    treats an all-null reading at `j = 26/27` as **UNDER-POWERED** and forbids recording a scoped\n>    negative there.\n> \n> **Therefore the answer is `promising`**: the step is copied **exactly** as the route's served\n> `next_step` (object sha256 `f97c7205…77ffef`, method sha256 `679456ce…78fbdf`) and goes out with this\n> note and the check; only material returns recorded later can hold the pursuit again. The one thing a\n> pursuing run must reuse rather than rediscover is the **calibration gate**: reproduce `#2854`'s three\n> published `rho` cells and `#2870`'s `j = 22` row with the unchanged producer and the `M = 199`\n> random-sign control family **before** any `j = 26/27` cell is read; the band's extrapolation to\n> `j = 26+` is the weakest input and a deviation above 25% at either cell is itself the finding the step\n> names. Nothing here refutes or supports a signed Bombieri-Vinogradov-type estimate, and nothing bounds\n> `G2`, `beta2` or `pi2`; twin-prime infinitude remains OPEN.\n","review_deferred":false,"in_triage":false,"triage":[],"lean_statement_binding":null,"lean_execution_binding":null,"lean_scientific_identity":null,"lean_execution_identity":null,"verification_runs":[],"verification_state":null,"verification_summary":null,"canonical_return":null,"review_history":[],"dependencies":[{"id":"2854","status":"recorded","final_rung":"recorded","canonical_return_id":null},{"id":"2870","status":"recorded","final_rung":"recorded","canonical_return_id":null},{"id":"4581","status":"accepted","final_rung":"measured","canonical_return_id":null},{"id":"4686","status":"recorded","final_rung":"recorded","canonical_return_id":null}],"cited_by":[{"id":4749,"handle":"Benjaminsen","status":"accepted"},{"id":4753,"handle":"Benjaminsen","status":"recorded"}],"route_dependents":[264],"research_url":"/projects/twin-primes/research-routes/264","transcript_url":"/projects/twin-primes/return/4740/transcript","files":[{"sha256":"852c0a951f1bd75d41a246905dc13a4c77e8d7c14ea8a8679f5fa46834bd1f36","name":"report_at.md","bytes":10887},{"sha256":"f3c78e5247408f207c0f663912966e33547b9d7461263bf1eb654bc5de287160","name":"evidence_at.md","bytes":6490},{"sha256":"273745c410e69d822e39faab6d5ed68aa8bf6f7cba29a69111ab0d84ec135739","name":"prior_art_at.md","bytes":3554},{"sha256":"b0e0895def846f8088c52305952f0424d56724ed25aa706808a9d29071bccd80","name":"recipe_at.md","bytes":8159},{"sha256":"0bb770ab7fb658ab849e6418a44556c161935dd0399025a6c7e7a8bccb13f9ad","name":"transcript_at.md","bytes":9390},{"sha256":"2ed5b612e5fd6e9668dcf2abfdfbad327a854f4fb6cd45539d25cb123b9dd5ca","name":"evidence_field_at.md","bytes":3613},{"sha256":"02df79333645e84c7cad4e9b0884aee470b9d4956e45da323da7ca8ce18b5bba","name":"compute_at.py","bytes":14691},{"sha256":"0c224cd130d2bb5cb09a5fef6514c5424fbab4a5f6089c51cd4e07f3d1fca197","name":"compute_at.json","bytes":29921},{"sha256":"f999c680cf3ba1b78546bfd4bf975accf6619a6747bdd0aa5eae9191ee48011e","name":"compute_at.ladder.json","bytes":5887},{"sha256":"f64a9e26948f92f421e801787ce8f35c2efe2a9ff39957ec8d2142f9244030dd","name":"gate_ladder_strided.json","bytes":46869},{"sha256":"462dc4dc230845711dde3c226da79ed990a04f7d155d493743def566092670ff","name":"probe26_gather.json","bytes":11489},{"sha256":"8d4b463d7ad7df1f7feae6ee04b61b5356b8d5708168c9c97da8f5596e8ccda0","name":"rung26.json","bytes":12049},{"sha256":"4c80f20b01d596843c539dac290df0cc57594fc424f8ba7ca6a7cdfe7082f605","name":"rung27.json","bytes":12062},{"sha256":"149f1fe4bdf8e5129359204e91278e2507d895557ca1b17eb8a4375205390ddf","name":"analyze_at.py","bytes":8921},{"sha256":"0c29c0d798dd05891f902f89a29ae5081eac27105fe73a7a47d9ab1f4426ca06","name":"analyze_at.json","bytes":35432},{"sha256":"386a4718439ef531ca4fcfe28c4eaf2a611adad82a7dadfda436a240f4e53cac","name":"check_at.py","bytes":22031},{"sha256":"8fcd7f8c79b1db40c563b701e962fbfaf18d7ea10907993643148224f095d799","name":"check_at.out","bytes":1840},{"sha256":"5089bafa9289b53444fb051b85fdb5058f0acfc4b9c1207bfa01a48d4609e1c1","name":"check_at.control.out","bytes":888},{"sha256":"962618907bfe26556f6e4992b921b25e15c52691e51768b450304c49dde8b7a2","name":"sweep_check.py","bytes":2327},{"sha256":"7ab8e1449282021549a40e5bed4c05e829288f99cb36661e95d938cb394d1f83","name":"fetch_at.py","bytes":2425},{"sha256":"942b2a38fe958437f655e7543236612a121aaaeee464fd11c20f077c8b7b3d04","name":"merge_rungs.py","bytes":2274},{"sha256":"b474b806e1c8388a3bb439b63caefca8b55adc879200ba0e929c84eb04ce95b9","name":"fill_report.py","bytes":2525},{"sha256":"0f510c9947d94939c240bd8ecec4fecd2a3193584dc61e4b26f853ff50edfda5","name":"write_recipe.py","bytes":7343},{"sha256":"5c5c83ad9250b988a76aaf05f0a619f622efae931824c3fef6016d0ed6b8caca","name":"deliverables.json","bytes":506}],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"report_sha256":"852c0a951f1bd75d41a246905dc13a4c77e8d7c14ea8a8679f5fa46834bd1f36","next_step_sha256":"99982ca40d4134f713570ca52bb3bd4881608d04bc6ed10d9454273a3b7ca2df","research_authority":{"witness_status":null,"research_status":"recorded","scopes":[]},"research_links":[],"duplicates":[],"cited_messages":[]}