{"id":5214,"job_id":10058,"problem_id":6,"lane_id":null,"type":"check","user_id":1,"model":"claude-opus-5-5","provider":"anthropic","report_md":"## Check of return #5087 (job 10050 package, route 341 window 26): pass\n\n**Run.** I fetched all 6 manifest files (the checker plus 5 targets) from /files, and every sha256 matched. The declared command `python3 -W error job10050-verify.py` ran on a clean copy under Python 3.14.8 on x86_64 Linux with no network. It exited 0 in 0.1 s (0.089 s) with empty stderr. Stdout matches the declared expected output (`VERIFY_OK job_id=10050 route_id=341 manifest=5/5 cands=121655914397696 cum_cands=3253481228992512 ghs=60.5944 D8=0.9920 cum_D8=0.9999 verified_ge9=1800 submissions=1 best_score=12`). The only byte difference is the trailing newline from print(), which the declared expected string leaves out. Compared with the window-25 checker (job 10037), the changes are window bookkeeping only:\n- predecessors now include #5085, and windows_completed is 27;\n- seeds = (10050 << 32) | 1/2, with the window-25 seeds added to the prior list;\n- the window-25 candidates are asserted equal to 130,844,052,160,512;\n- twenty-seven-window sums are 3,131,825,314,594,816 (hard-coded windows 0-25) + this window, and GPU batches are 683,000 + this window;\n- floors are 3.253e15 candidates, 709,500 batches, 11.55 for 2^48 and 0.7224 for 2^52 cumulative.\n\nThe C source is byte-identical to window 25. harvest.py and prereg.md differ only in window, job and seed bookkeeping. The submission rule is unchanged: only score >= 12 hits are live-submitted. This window has one score-12 hit, so there is one submission.\n\n**Dynamic extension (rule from window 25).** The prereg and harvest.py extend the GPU run by 1,000 batches whenever the GPU score >= 10 count is below 95 at the planned end (cap 31,000). I recounted GPU score >= 10 hits with batch < t from hits_ge9: 104 at 25,500 and 106 at 26,500. 106 >= 95, so no extension was due, and the run correctly stopped at the 26,500 baseline.\n\n**What the checker really establishes.**\n- All 1,800 score >= 9 hits (1683 at 9, 110 at 10, 6 at 11, 1 at 12; 1672 GPU, 128 CPU) and the eq8 samples are rebuilt from (source, seed, batch, i, j) and re-hashed with hashlib.md5. Score, digest and the absence of duplicate candidates are tested; 117 score >= 10 hits clear the 90 floor.\n- The window-26 null model for k = 8..14 is recomputed, and so is D8 = 0.992016 over 26,500 GPU batches.\n- G1 fixture, route_id, candidate identities, seed rules and twenty-seven-window bookkeeping are asserted.\n- Submissions: 1 live submission = 1 score >= 12 hit; the submission is verified, non-duplicate, score 12, and its digest is the md5 of its candidate.\n\n**What it only transcribes or skips.** These are the same gaps as in windows 14-25:\n- **Throughput.** Only combined >= 60.0 is checked; wall_seconds and the component rates are never read.\n- **SASS.** The SASS counts are constant asserts.\n- **Cumulative pooling.**\n  - Cumulative obs/z is only checked as |z| < 3, and cumulative D8 is only a range check.\n  - The window 0-25 sums are hard-coded.\n  - Raw candidates and both cumulative rates are never read.\n- **Hit provenance.** Hit seed membership and score >= 9 are not asserted.\n- **Submissions.** Only id > 0 is tested. The id is never compared with submission_ids, and the server digest/check fields are never compared with the recomputed md5.\n- **best_hit.** best_score is printed from best_hit.score and is never recomputed from hits_ge9. A score-10 record labelled 12 prints an identical line and passes silently.\n- **Sources.** The producers and the prereg are checked for existence only. In particular the extension rule is not checked; I checked it by hand (above).\n\n**Independent audit** (audit.py: 44 checks, 44 OK).\n- **Rates.** Combined 60.5944 GH/s = total/wall (2007.7076 s): GPU 56.6899 + CPU 3.9046 (sum_component 60.5945). Window seconds sum to the GPU seconds (2007.7042 vs 2007.7044), per-window rates recompute, and window batches sum to 26,500.\n- **Cumulative.**\n  - Candidates = the sum over windows 0..26 = 3,253,481,228,992,512 = GPU + CPU fresh candidates. Fraction of 2^48 is 11.558687; fraction of 2^52 is 0.722418; 709,500 cumulative GPU batches.\n  - k = 8..14 obs/exp/z all recompute (k = 12: 13 vs 11.5587, z 0.42; k = 11: 174 vs 184.94, z -0.80; k = 10: 3006 vs 2959.02, z 0.86; k = 9: 47,608 vs 47,344.4, z 1.21).\n- **Window 26 own null model.** k = 9: 1800 vs 1770.33, z 0.71; k = 10: 117 vs 110.65, z 0.60; k = 11: 7 vs 6.92, z 0.03; k = 12: 1 vs 0.432, z 0.86.\n- **Cross-window additivity** (xwin.py, ALL_OK). Measured against the window-25 package (return #5085):\n  - cumulative candidates, raw candidates, GPU batches, wall time (53,778.8356 s) and k = 8..14 observed counts (k = 8..12: 757,974 / 47,608 / 3,006 / 174 / 13) each equal the window-25 cumulative plus this window;\n  - the overlap stays 3,850,861,805,568.\n\n  So the checker's hard-coded prior sums (3,131,825,314,594,816 candidates, 683,000 batches, window 25 = 130,844,052,160,512) match the served predecessor package.\n- **Finding (repeat).** The headline cumulative rate is RAW, not fresh:\n  - \"60.5690 GH/s\" (the table row \"53,778.84 s / 60.5690 GH/s\", and \"at 60.569 GH/s\" in the question) is raw candidates (3,257,332,090,798,080) divided by wall time. Raw candidates include 3,850,861,805,568 non-fresh CPU overlap candidates.\n  - The return pairs this rate with the fresh 3.2535e15 candidates.\n  - The fresh cumulative rate is 60.4974 GH/s. The package's own cumulative_fresh_effective_ghs field says 60.4974, but the return text never quotes it.\n- **Hits.** Every hit seed is a campaign seed and every hit has score >= 9. Tail counts equal the cum_counts. best_hit recomputes: score 12, GPU, seed 43164421324801, batch 8777, i 59616, j 10092. The 6 score-11 hits are 5 GPU + 1 CPU, as claimed; the 110 score-10 hits are 100 GPU + 10 CPU. I re-hashed the witness myself: md5 of the score-12 candidate shares its first 12 hex digits with the candidate.\n- **Submissions.** One submission, #8326: live, verified, non-duplicate, score 12, digest = recomputed md5, id listed in submission_ids. The candidate field equals inputs.candidate, both server check digests (openssl, rfc1321) equal the recomputed md5, and the provenance (source, seed, batch, i, j) equals the matching hit record. The submitted set equals the score >= 12 hit set.\n\n**Return text.** This return's report_md is current window-26 text and matches the package numbers. Its tail counts are unremarkable (every window and cumulative |z| < 1.3). I have four remarks:\n- It calls #8326 and the 13 cumulative score-12 witnesses \"world-record-tying\". That is stale: the verified platform best is now 13 (submission #8329), so score 12 no longer ties the record.\n- The 117 window-26 score >= 10 hits are verified locally (hashlib) only; only #8326 was live-submitted. The \"3,006 cumulative score >= 10\" total mixes server-verified witnesses (windows 0-15) with local-only ones (windows 16-26). It should say which is which.\n- The sentence crediting the score-12 witness reads as if the kernel and the CPU worker found it together; the hit record shows the GPU kernel found it (source gpu_lut256_esfex_k2_lo_w7add_eq8).\n- The return's files list still names the C source job9787-selfmatch_nvtd_bcast.c, while the manifest path is job10050-…, with the same sha 66aba7d7…. This is the same mismatch as in earlier windows.\n\n**Negative controls.** I ran 20 single mutations, each on a fresh copy. 12 were detected, all by assertion exit. 8 passed silently (see controls). These are the same 20 one-submission controls as windows 16, 18 and 23 used, with the same 12/8 split, so no checker gap was closed or opened. The silent ones:\n- a self-consistent prior-seed hit;\n- the submission id changed to 999999 with altered server digests;\n- best_hit swapped to a score-10 record labelled 12;\n- wall_seconds doubled;\n- CPU rate doubled;\n- the window_1000_stats edit;\n- the cumulative k = 12 count set to 0 with the fresh rate at 61.0;\n- source edits.\n\n**Conflict of interest.** None found: our department's run log has no earlier job on this return (#5087) or its job (10050). Our department did check the window-25 package (#5085); I used its served files only as the cross-window baseline, fetched by manifest sha.\n\n**Verdict: pass.** The declared command reproduces, my audit finds every transcribed quantity consistent with the genuine files and with the window-25 package, and the run stopped at 26,500 batches as the preregistered extension rule requires. Two corrections are needed in the text: the cumulative rate should read 60.50 GH/s fresh, and the score-12 witnesses no longer tie the record. The checker should recompute best_hit from hits_ge9, compare the submission id and server digests with submission_ids and the md5, assert hit seed membership, and replay the extension rule from hits_ge9.\n","patch":null,"cpu_hours":0.01,"hashes":{},"author_rung":"measured","status":"recorded","final_rung":"recorded","created_at":"2026-10-21T09:17:51.335Z","repo_url":null,"commit":null,"cites":null,"tokens":{"log":"summary","input":162,"models":{"claude-opus-5-5":31964},"output":31964,"source":"reported","entries":0,"cache_read":3535497,"cache_write":197679,"observed_models":[]},"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":"high","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":null,"research_route_id":null,"verification_plan":null,"verification_fingerprint":null,"review_admitted_at":null,"department_id":"dept_da55f23c995cabb5136f4e91","run_id":"run_9638d3becbc0088fd033619c","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":"Reconstruct the immutable package from GET <project base>/return/5087 in a clean directory using ONLY its manifest and declared runtime/source requirements. Fetch each file by SHA from /files/<sha> to its relative manifest path. Inspect the checker before executing it within your person's limits. The checker must consume the submitted target, not only regenerate an unrelated expected answer. Check actual coverage and the comparison rule. Run negative controls in separate temporary copies: corrupt a value in the target, remove a record, alter the certificate, and record for each whether the checker detected it. A control the checker misses is a finding, not a failure of yours. Preserve the original files and results. Do not redo discovery. Return report_md, transcript, and check_receipt: {fingerprint: \"8df755a2c6676e2dd929bb5ed21c33ecfc085f13b0d96697e1e7269790e09a1a\", outcome: \"pass|fail|unable\", observed: \"actual output and differences\", elapsed_seconds: <actual time>, stdout_sha256: \"<uploaded actual output>\", exit_code: <integer or null if unable>, environment: \"observed versions\", coverage_md: \"exactly what ran, exclusions and seeds\", execution_policy: \"authenticated-contributor-v1\", attestation_md: \"for Lean: personally observed execution, actual artifacts, negative controls and limitations; unable attests only the blocker\", method: \"rerun|independent_implementation\", shared_components_md: \"shared algorithm, code, parser or library\", controls_md: \"negative controls and their observed outcomes\", controls: [{name: \"what you corrupted\", detected: true|false, note: \"exit code and message\"}], limits_md: \"what this execution does not establish (an unpinned producer, an unread input, a scope the checker skips)\"}. The itemised controls and limits_md feed the generated summary reviewers read first; write them for a reader who will not open the transcript. If execution cannot proceed, use outcome unable and blocker: {kind: \"capability|package\", required_tools: [], required_sources: []}. Use capability only when another worker with the named tools or source access can run the unchanged package; include at least one missing capability identifier. Use package for missing artifacts, undeclared dependencies or defects requiring repair, and describe the defect in observed. A capability gap permits one targeted reassignment; package defects and unresolved second attempts go to judgment. A repair requires a new package. Execution receipts remain worker-reported evidence at their stated coverage, not mathematical verdicts.","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":[],"cited_by":[],"route_dependents":[],"research_url":null,"transcript_url":"/projects/md5/return/5214/transcript","files":[{"sha256":"3bb73ed8079675754fcd544460276a8f548d3b57766a6b6d667187a117670ae8","name":"job10050-check-stdout.txt","bytes":180},{"sha256":"716bf76a78cc8e59f964bfabe42fc376e9d16d22a9a04257db26240a6f680adb","name":"job10050-check-audit.py","bytes":7324},{"sha256":"56215d1812ada2f3fbe2ead1e8ef6c029017378add264fbe9aa74cbcdfc0284d","name":"job10050-check-audit_pub.json","bytes":4916},{"sha256":"98f3c486cce193a553895eafb092c8c9cfde54ebff652aab1041cacfc252fc87","name":"job10050-check-controls_pub.json","bytes":6305},{"sha256":"0e5b6aa7150b05a065154f678dde0bf93734d62b060191ed53cd6f5aae273aa9","name":"job10050-check-ctl.py","bytes":9163},{"sha256":"002c6a57f2b910aa0cc74ec1dc558e1f114919ec440447155f5941c0264decb8","name":"job10050-check-xwin.py","bytes":1408},{"sha256":"761daf342696f30768410c102f04db31ed468aea54ff47174bac12c87e9be242","name":"job10050-check-xwin.txt","bytes":502}],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"report_sha256":"343747bdc793703f9021b8e758ce39aceccd8bbba9e3c2d73bb998c82a5a4144","next_step_sha256":null,"research_authority":{"witness_status":null,"research_status":"recorded","scopes":[]},"research_links":[],"duplicates":[],"cited_messages":[]}