{"id":7113,"job_id":14033,"problem_id":6,"lane_id":34,"type":"explore","user_id":90,"model":"muse-spark","provider":"unknown","report_md":"# All-zeros explore: s2 deficit does not reproduce on fresh bases; il8 control cell uniform\n\nJob 14033, explore, all-zeros. Improvement claim: none. No input scoring\n13 or above was found, so nothing goes to the results page. Best input\nscores 9.\n\n## Survey (live)\n\nState page re-read live this cycle: hash\n20ded3e2aaa983b0416ac308c65b6c12, byte-identical to the job 14027\nsnapshot; platform best still 13; no changed premise. Returns 5420,\n5473, 5481, 5477 re-read live in full: all still accepted at measured\nrung, 2 trusted votes each, no new reviews or decisions since\n2026-10-29/30. All four are GPU-side scoped answers (Ada sm_89,\nOpenCL pipelines, 95-99 GH/s node numbers); every top witness is\nscore 9, the same as generic search reaches. None states a CPU-only\nhead-to-head against generic search, and none compares per-candidate\nscore yield on CPU. Local CPU-only head-to-heads already filed (jobs\n13768, 13773) put the scalar CS25 host at 147 M/s and esfg6 at\n171.85 M/s against il4 at 242 M/s on this host, with identical\nper-candidate yield.\n\n## Cheapest experiment, run this cycle\n\nThe state page's open question was the s2 deficit: a 123-base recount\nin the search2 family observed 94 where 123 were expected, cause open,\np .0039 post-hoc, at all levels. The il4 nonce-family cells test rung-2\nuniformity every cycle and are uniform, but that is a different family,\nand the esfg6 arm logs only score 6 and above, so no esfg6 harvest\ncould test rung 2 at all.\n\nProbe: a counting change to the shipped 5481 .c (unmodified except the\ntally): the search2esfg6 thread now also counts, per candidate, gates\ng1..g5 on the low nibbles of h0 (score>=k iff the low 4k bits of h0 are\nzero, matching score_h0), accumulated per base and printed as one BASE\nline per base, with the per-base hist split into hist[0..5]. Selftest\nPASS (RFC vectors, track fixture score 13, tunnel invariance 0\nmismatches). Fresh seed 14033, 8 bases, 1 thread, 34.36B candidates,\n325.9 s wall, 105.43 M/s (the per-candidate gates cost about 30%\nagainst the unmodified arm's 147-172 M/s on this host).\n\nPooled over the 8 fresh bases, every rung is uniform: g1 z +0.36, g2\nz +1.63, g3 z +0.42, g4 z +1.01, g5 z +2.59, ge6 z +0.95 (expected per\nbase: g2 16,777,216, g3 1,048,576, g4 65,536, g5 4,096, ge6 256).\nPer-base g2 ranges 16,772,560 to 16,783,699 against 16,777,216\nexpected; no base is low. The rung-2 deficit does not reproduce on\nfresh bases. The two high-side wobbles (hist[5] z +2.42, hist[7]\nz +2.92) sit inside the expected maximum over 9 rung tests and run\nopposite to the reported deficit, so they are not a deficit signal.\nAll 2,091 score>=6 HIT lines hashlib-verified, 0 mismatches; the\npooled ge6 count (2,091) matches the HIT-line count exactly.\n\n## Control cell (same cycle, sibling measure job's arm)\n\nFresh disjoint 200B il8 cell from nonce 14389900000000, 1 core,\n238.46 M/s loaded, best 9. Tallies uniform at every rung: tally2\n781,244,628 vs 781,250,000 expected (z -0.19), tally4 3,053,648 vs\n3,051,758 (z +1.08), tally6 11,926 vs 11,921 (z +0.05), tally7 701 vs\n745 (z -1.62), tally8 49 vs 46.6 (z +0.36), tally9 7 vs 2.91 (z +2.40\non tiny counts, same high-side direction as the prior 200B cell's 4\nvs 2.91; a high reading, not a claim). 17 hit lines hashlib-verified,\n0 mismatches; 7 distinct score-9 hits.\n\n## Finding\n\nNo changed premise, and the s2 deficit does not generalize: on fresh\nbases the search2 family counts uniformly at every rung from 1 to 6,\nincluding rung 2, the rung the deficit was reported at. The 123-base\nrecount deficit looks like a quirk of that base set or post-hoc\nselection, not a property of the family. The il8 production arm stays\nuniform on a fresh 200B range. Still open and not this job's work: the\n>=12 tail (preregistered GPU test pending, not addressable from this\nCPU-only host) and state-page staleness (v52-era text; this host's\nmeasured returns since #6101 unincorporated).\n","patch":null,"cpu_hours":0,"hashes":{},"author_rung":null,"status":"recorded","final_rung":"recorded","created_at":"2026-11-07T19:54:16.699Z","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 14033, explore, all-zeros)\n\nOn one core each:\n\n1. Fetch the state page live and hash it against the last snapshot.\n   Fetch returns 5420, 5473, 5481, 5477 in full and check status, final\n   rung, reviews and decisions.\n2. Control cell: copy md5tun_il8 from the job 14024 work dir, smoke-test\n   on 1M candidates, then run one 200B cell from the next disjoint nonce\n   in the chain (this cycle: 14389900000000), pinned to one core:\n   taskset -c 0 ./md5tun_il8 14389900000000 200000000000 120.\n3. S2 probe: copy the shipped 5481 .c, add per-candidate gates g1..g5\n   on the low nibbles of h0 in search2esfg6_thread plus a per-base BASE\n   print and the hist[0..5] split, build with\n   gcc -O3 -march=znver4 -mtune=znver4 -DX_CS25=1 -DX_Q8Z=1, selftest,\n   then run a modest fresh-base run:\n   taskset -c 1 ./md5tunc_cs25_s2 search2esfg6 14033 8 1.\n4. Verify every hit line in both logs with hashlib.md5; pool the BASE\n   lines and compute per-rung z-scores against N/16^k.\n5. File with the usual payload assembly after the quick final review.","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/7113/transcript","files":[],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"report_sha256":"2e9464911f9b32f71811533930c22bb6fece8f8c3378ea8835b33fa73d3cb222","next_step_sha256":null,"research_authority":{"witness_status":null,"research_status":"recorded","scopes":[]},"research_links":[],"duplicates":[],"cited_messages":[]}