{"id":6986,"job_id":13772,"problem_id":6,"lane_id":34,"type":"measure","user_id":90,"model":"muse-spark","provider":"unknown","report_md":"# All zeros: AVX-512 esf arm still trails generic il4 on a fresh base; 60B tally cell uniform, best 9\n\nJob 13772, measure, all-zeros. Improvement claim: none. No input\nscoring 13 or above was found, so nothing goes to the results page.\nBest input scores 9.\n\n## Hypothesis\n\nThe AVX-512 Q9-tunnel esf arm (md5tun_avx512_esf, 138.95 M/s solo in\njob 13092) closes the gap to generic il4 on a fresh disjoint base, and\nits per-candidate yield matches uniform. Motivation: the published\n5.3 GH/s CPU aggregate over 8 threads (~664 M/s/thread) is 4.5x the\nscalar CS25 host rate measured last cycle (147 M/s), and the open\nquestion was whether a vectorized esf arm explains it. If the esf arm\nbeat il4 here, the tunnel route would explain part of the aggregate;\nif it still trails, the aggregate is not explained by any arm measured\non this VM.\n\n## Experiment\n\nRebuilt both binaries from source (gcc -O3 -march=znver4\n-mtune=znver4), selftests PASS. One core for all work in this job; a\nparallel job ran on the other core. A VM reset happened at cycle start,\nbefore any cell ran; no reset occurred during the runs.\n\n1. AVX-512 esf arm: fresh 20B cell from nonce 20000000000 (bases 4 and\n   up, disjoint from the 13092 benchmark base), log interval 60 s.\n2. Generic il4 tally baseline: fresh disjoint 60B cell from nonce\n   12089840000000, log interval 60 s.\n\n## Results\n\nESF cell: 20,000,000,000 candidates at 101.07 M/s overall (solo rate\nin 13092 was 138.95 M/s; the parallel job loaded the other core for\nmost of this run). Best 8:\ninput 8d0299d8be5f26eb786e0989139d38723b5861a2049fe89d2e7c52d6822dec31d96653d3401aed0f5422ad8f37a1d2d579e54913,\ndigest 00000000b435459663015d19fe2ab212, 52 bytes. All 9 new-best hits\nhashlib-verified, 0 mismatches.\n\nIL4 tally cell: 60,000,000,000 candidates at 228.94 M/s overall (about\n260 M/s while the other core was idle, lower while the parallel job\nloaded it). Best 9:\ninput b66bff74978ef937a44ec1d235371c56a1747bc50e754e57f4243bf09c67fe727a11b83cb4e392bf1be806876406948714998e8f,\ndigest 0000000007ac7e954dbfb29b8b633888, 52 bytes. All 9 new-best hits\nhashlib-verified, 0 mismatches.\n\nTallies for the il4 cell (observed / expected / z): tally2 234,359,593\n/ 234,375,000 / -1.01; tally4 914,997 / 915,527 / -0.55; tally6 3,605 /\n3,576 / +0.48; tally7 221 / 223.5 / -0.17; tally8 15 / 14.0 / +0.27.\nEvery rung within 1.1 sd of uniform.\n\nCounted total now about 800B: tally7 3015 vs 2980.2 expected (+0.64\nsd), tally8 197 vs 186.3 (+0.79 sd). No bias at any rung, no drift\nagainst earlier cells.\n\n## Verdict\n\nHypothesis refuted on rate: the vectorized esf arm runs at 101 M/s\nhere against il4 at 229 M/s in the same cycle (ratio 0.44), and 139\nM/s solo against il4 at about 242 M/s solo. The tunnel route does not\nbeat generic search on this VM, and the 5.3 GH/s aggregate remains\nunexplained by any arm measured here. Yield is uniform in both arms.\nThe fastest local arm stays the production il4 binary.\n","patch":null,"cpu_hours":0,"hashes":{},"author_rung":null,"status":"pending","final_rung":null,"created_at":"2026-11-05T16:51:04.479Z","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\n\nTo repeat, on one core:\n\n1. Rebuild md5tun_avx512_esf and md5tun_tally from source with\n   gcc -O3 -march=znver4 -mtune=znver4; run both selftests.\n2. Run a 20-billion-candidate esf cell from the next fresh nonce\n   (this cycle: 20000000000), log interval 60 s.\n3. Run a 60-billion-candidate il4 tally cell from the next disjoint\n   nonce (this cycle: 12089840000000), log interval 60 s.\n4. Verify every [new best] line in both logs with hashlib.md5\n   (digest and true leading-zero score must match the printed values).\n5. Compute per-rung z-scores against N/2^(4k) expectations and add\n   the tallies to the running counted totals; compare the esf and il4\n   rates measured in the same cycle.","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":"2026-11-05T16:51:04.479Z","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":"Study what makes the first output word of MD5 small, and use it to reach more leading zeros than generic search would with the same compute. Ideas to test: freedom from extra message blocks, neutral bits and message modification from collision attacks applied to the output instead of a difference, early abort on the final additions. Start from the algorithm, not the search. Read the track's state of the art in this brief (the fastest known way per machine, what worked, the dead ends) and research/QUESTIONS.md, then state one hypothesis about MD5's structure that would make this track cheaper than generic search, and why you expect it. Test it with the smallest experiment that could refute it, against a measured baseline on the same machine. Submit the best candidates the experiment produced. The report is a finding: the hypothesis, the experiment, what it showed about MD5 (positive or negative, with numbers), and what the next run should try. Send what you learned as a state_entry with your return (the method, the exact hardware, the measured throughput, the best reached), so the track's state of the art carries it. If the run used only a known tool or plain search, report it as a baseline measurement.","review_deferred":false,"in_triage":true,"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/6986/transcript","files":[],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"report_sha256":"e3fe8e4f7184ff0a60fa497c53575afd15e823cfe6b9db7c4064a1732a64fc38","next_step_sha256":null,"research_authority":{"witness_status":null,"research_status":"pending","scopes":[]},"research_links":[],"duplicates":[],"cited_messages":[]}