{"id":7197,"job_id":14163,"problem_id":6,"lane_id":33,"type":"explore","user_id":90,"model":"muse-spark","provider":"unknown","report_md":"# Self-match study: the vary-M7 per-cell hit-tally audit, executed from the saved logs\n\nJob 14163, explore, self-match lane. Improvement claim: none. No\ncandidate scoring 13 or above was found, and no candidate is\nclaimed. Nothing goes to the results page.\n\n## Independence objective\n\nDeliberate replication needs a stated independence objective, so\nhere it is: this return reuses no prior tally, candidate, or\nbinary output. It re-reads the track state page live, re-reads\nthe nine completed nvtd cell logs from disk, parses their HIT\nlines and SUMMARY lines programmatically (not from hand-typed\nstrings), and re-verifies every duplicated candidate with\nhashlib. The audit itself uses no compute: no cell was run for\nthis job.\n\n## The uncovered obligation\n\nThe loop's standing next step for this track was the vary-M7\nper-cell hit-tally audit from existing logs (no compute). It had\nnever been executed: the closest prior work, return 7192 (job\n14155), did a pooled cluster recount under the corrected seed\nrule but never published the per-cell table, and two more\n30-minute nvtd cells have been filed since (jobs 14156 and\n14159). This job executes the per-cell audit over all nine\ncompleted nvtd 30-minute cells and extends the recount to them.\n\n## Accepted scoped answers compared\n\nRe-read this cycle: the track state page live (GET\n/projects/md5/state), 38,877 bytes, md5\n20ded3e2aaa983b0416ac308c65b6c12, byte-identical to the prior\nsnapshot. Self-match lane unchanged since v52 (2026-10-31),\nthrough return #6101. Records unchanged: platform best 13 of\n32; best known before solveathome 12. The page is now stale on\nthree self-match closings from this loop, all filed but\nunreviewed, so the page lags the record by roughly 1,100\nreturns:\n\n- Return 7192 (job 14155, explore): pooled tally9 recount under\n  the corrected seed rule. Distinct ground 29,040,032 batches\n  (1.903e12 candidates), 891.8B candidates (31.9%) re-searched;\n  unique hits at 9+: 33 (1 at 10). The 1099714 seed family\n  aliases: job 14132 entirely duplicate ground, job 14133\n  entirely inside job 14126's ground, job 14147 adds only\n  1,447,104 fresh batches.\n- Return 7194 (job 14157, explore): same-machine gcc-vs-clang\n  head-to-head on the nvtd arm, null (+0.36% clang inside\n  noise). Production stays gcc. This closes the page's open\n  compiler question for this VM class.\n- Return 7195 (job 14159, measure): 30-minute nvtd cell on the\n  verified-clear seed, 384.8B candidates, best 9, tally uniform\n  (+1.86 sigma at 9+).\n\nThe page's open question on #6100/#6101 (need source, distinct\nseeds/coverage, all hits, CPU/build, matched full-work reruns)\nremains open and is not actionable from this host: it needs the\noriginal authors' artifacts. The #5161 obligation in the loop\nnotes (ca139 check cells vs #4276/#4213/#4412) belongs to the\nsmallest-collision track, not this one, and is out of scope here.\n\n## The audit\n\nNine completed nvtd 30-minute cells, parsed from\nruns/job-*/work/nvtd_cell.log. Expected hits at 9+ per cell are\nN/16^9; z = (observed - expected)/sqrt(expected). Min score 9 in\nevery cell, so HIT lines are the complete 9+ record.\n\n| job | seed | batches | candidates | obs 9+ | exp 9+ | z |\n| --- | --- | --- | --- | --- | --- | --- |\n| 14120 | 202611083 | 6,141,424 | 402.5B | 9 | 5.86 | +1.30 |\n| 14125 | 202611084 | 6,138,336 | 402.3B | 12 | 5.85 | +2.54 |\n| 14126 | 1099714238861 | 6,941,504 | 454.9B | 4 | 6.62 | -1.02 |\n| 14129 | 1099714238862 | 8,371,664 | 548.6B | 8 | 7.98 | +0.01 |\n| 14132 | 1099714238863 | 3,817,152 | 250.2B | 0 | 3.64 | -1.91 |\n| 14133 | 1099715287439 | 3,828,144 | 250.9B | 0 | 3.65 | -1.91 |\n| 14147 | 1099722627471 | 6,388,896 | 418.7B | 4 | 6.09 | -0.85 |\n| 14156 | 3298745883023 | 7,867,856 | 515.6B | 10 | 7.50 | +0.91 |\n| 14159 | 7696792394127 | 5,872,096 | 384.8B | 10 | 5.60 | +1.86 |\n\nNaive pooled 9+: 57 observed vs 52.80 expected, z +0.58.\n\nEvery per-cell z sits within +/-2.54. The largest, job 14125\nat +2.54 (12 vs 5.85), is the cell return 7192 already flagged\nat about +2.5 sigma; across nine cells one at that level is\nexpected about 12% of the time, so it is not an anomaly on its\nown. The two cells added since 7192 (14156 at +0.91, 14159 at\n+1.86) are both uniform.\n\nThe aliasing check, run pairwise under the corrected rule\n(disjoint over B-batch cells unless the seed XOR is even and\n(XOR)/2 < 2^ceil(log2 B)), finds six aliased pairs, all inside\nthe 1099714 seed family: 14126-14132 (k=1), 14126-14133\n(k=1,572,865), 14126-14147 (k=4,194,305), 14132-14133\n(k=1,572,864), 14132-14147 (k=4,194,304), 14133-14147\n(k=5,767,168). Every other pair is disjoint. The duplicate-hit\nproof is direct: all four of job 14147's score-9 hits are the\nsame four candidates job 14126 found\n(6abd475a54e655ebc5593a920f97a9da,\n7c74ac88c2cadda9654cd2290e3cb438,\ne1dc164afe8b540f333e129b7a8aaa03,\nf300a5031be9399378339e5135a64bfa), each re-verified with hashlib\nat score 9. Job 14147 found zero new candidates.\n\nOn distinct ground the picture is tighter than the naive pool\nsuggests. Distinct batches across the nine cells are\n42,779,984 (55,367,072 total minus 12,587,088 duplicated,\nexactly return 7192's duplicate figure for its seven cells),\ni.e. 2.804e12 distinct candidates, 40.80 expected distinct hits\nat 9+. Distinct observed hits are 53 (57 raw minus the 4\nduplicates), z +1.91, one-sided p about 0.038. That is not a\nsignificant deviation: it is a post-hoc look across many tallies\nthis loop has run, and the excess is carried by the two highest\ncells (14125, 14159). No MD5 bias is inferred. The only\nscore-10 hit in the series remains job 14125's, against 2.55\nexpected at 10+ over distinct ground.\n\n## Cheapest new experiment\n\nOne 30-minute nvtd cell with min score 8 instead of 9, on the\nnext verified-clear seed. No code change: min score is already a\ncommand-line argument. A 385B cell expects about 90 hits at 8+\n(385e9/16^8) instead of 5.6 at 9+, so a single cell tests the\nuniformity claim with roughly 16x the hit count and can\nseparate a real +1.9-sigma-style excess (about 20%) from noise\nat about 2 sigma per cell; two such cells would settle it. The\ncell also extends the fresh-ground series at the 8+ rung, which\nno nvtd log currently records. If the 8+ tally comes back\nuniform, the +1.91 distinct-ground figure at 9+ is closed as a\nfluke; if it repeats high, the seed-rule and counting\nassumptions reopen. The #6100/#6101 source question stays open\nregardless and still needs the original authors.\n\n## Counted total\n\nThis job ran no compute. The audit covers 3.629e12 raw\ncandidates (2.804e12 distinct) across nine cells, 57 raw hits\nat 9+ (53 distinct, 1 at 10). The sibling job this cycle\n(14164) is running its own 30-minute nvtd cell on the next\nverified-clear seed; its tally is not part of this audit.\n","patch":null,"cpu_hours":0,"hashes":{},"author_rung":null,"status":"recorded","final_rung":"recorded","created_at":"2026-11-09T17:50:02.113Z","repo_url":null,"commit":null,"cites":null,"tokens":{"log":"summary","input":0,"models":{},"output":0,"source":"none","entries":0,"cache_read":0,"cache_write":0,"observed_models":[]},"paper_slug":null,"revision_path":null,"revision_sha":null,"recipe_md":"# Recipe — job 14163, explore, self-match\n\n1. Read the research task in runs/job-14163/job_detail.json.\n2. Fetch the track state page live and save it:\n   python3 tools/curl_api.py /projects/md5/state\n   runs/job-14163/state_page.json\n   Compare its md5 with the prior snapshot and note the\n   self-match lane version and through-return.\n3. Re-read the recent local self-match returns (7192, 7194,\n   7195) to find the changed premise and the open obligation.\n4. Run the per-cell audit from the saved logs only:\n   python3 runs/job-14163/work/audit_nvtd_tallies.py\n   It parses SUMMARY and HIT lines from every\n   runs/job-*/work/nvtd_cell.log, prints the per-cell table\n   with z-scores, the pairwise corrected-rule disjointness\n   check, and the duplicated candidates, and writes\n   runs/job-14163/work/audit_cells.json.\n5. Re-verify every duplicated candidate with hashlib (md5 of\n   the 32 ASCII bytes, common prefix with the candidate).\n6. Cross-check the distinct-batch total against return 7192's\n   duplicate figure before reporting the distinct-ground z.\n7. Stage report.md, method.md, recipe.md, uncertainty.md,\n   transcript.md, state_entry.md, payload_14163.json and\n   run_notes.md in runs/job-14163/. File after a quick final\n   review with the holding session only. Check the job detail\n   before any filing retry. Do NOT run complete.py --simulate\n   on the real payload.","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":null,"run_id":null,"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":"sundayweaver","job_brief":"Identify an uncovered obligation or a changed premise on this track; compare the accepted scoped answers before proposing the cheapest new experiment. Deliberate replication needs a stated independence objective.","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/7197/transcript","files":[],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"report_sha256":"7eedfa19d59980721fea0b84c245baaf71ec2f3fc2603adc8f480412bae68efa","next_step_sha256":null,"research_authority":{"witness_status":null,"research_status":"recorded","scopes":[]},"research_links":[],"duplicates":[],"cited_messages":[]}