{"id":6823,"job_id":13557,"problem_id":6,"lane_id":34,"type":"measure","user_id":90,"model":"muse-spark","provider":"unknown","report_md":"# Measure: sustained il4 rate and score distribution on a fresh 3.1e11 cell; best 10, uniform model survives\n\nJob 13557, measure, all-zeros lane. Improvement claim: none. No\ncandidate scoring 13 or above exists, so nothing goes to the results\npage. All finds below are recorded in this return and the tally\nonly.\n\n## Hypothesis\n\nStated before measuring: on one pinned CPU core, the interleave-4\nAVX-512 tunnel sustains a rate near its solo rate (about 240 M/s),\nand the score distribution of a fresh disjoint cell follows the\nuniform model: for 310,000,000,000 candidates the expected max is\nlog16(3.1e11) = about 9.5, and score 10 or better is expected\n0.28 times, so seeing at least one such hit has probability about\n25 percent under the model. A sustained rate far below solo, or a\nscore distribution far from these counts, would refute the\nbaseline-measurement framing and point to either machine\ninterference or structure in the search.\n\n## Experiment\n\nVerified the production binary (tools/md5tun_avx512_il4,\ninterleave-4) with its selftest (SELFTEST PASS: vector vs scalar\nagreement over 100,032 candidates) before burning compute. Ran one\nclean single-segment pilot cell of 310,000,000,000 candidates\nstarting at nonce 42100000000000, pinned to a single core with\ntaskset, log interval 60 s. The range is disjoint from prior cells\n(#13553 used 41000000000000 plus 3.1e11, #13552 used\n41500000000000 plus 3.1e11) and avoids the\n42000000000000-42014400000000 bench range. Every top hit at score\n8 or above was recomputed with an independent hash implementation\n(Python hashlib).\n\nA VM reset interrupted the first attempt at about 840 s into the\ncell (no done line, process gone, uptime reset). That partial run\nwas discarded, not resumed mid-range, because the exact completed\ncandidate count was not recoverable from 60 s log lines; the full\ncell was re-run from the same start nonce after the reset, and the\nresults below are from the complete re-run. The partial run had\nindependently reached the same score-10 hit before being killed,\nwhich corroborates it.\n\n## Results\n\n- Cell: 310,000,000,000 candidates, overall 233.40 M/s, best 10.\n  Early rate about 241 M/s, easing to about 233 M/s late in the\n  run, a dip of roughly 3 percent, consistent with a second worker\n  starting on the other core late in the run. The sustained rate\n  is within about 3 percent of the solo rate, so the rate half of\n  the hypothesis survives.\n- Best candidate, verified by recompute: input\n  8cf08c9358b27e3163686f153634cd3440a6ca3c202a3db0a73bdeecdff3132d15c0631931af2c087723ea04f5e0b2b3cf83aa22\n  (52 bytes) -> digest 00000000009a3bd5586d65d562e5a4a9, score 10.\n- Also verified: score 8 input\n  4688e57901dbfe4c36e411dcd97b5809e97b7b004a487164b5057103cffb6be91f8cc876238301d809031e534b59040159b4301d\n  -> digest 0000000033c931fafdf5d013d217e380. No score-9 hit was\n  logged because the best jumped from 8 straight to 10; the\n  searcher logs only record-breaking hits, so later score-10\n  hits, if any, would not have been logged either. The score-10\n  count is therefore at least one, not exactly one.\n- Distribution check: best 10 vs about 9.5 expected max; at least\n  one score-10-or-better hit vs 0.28 expected, an event with\n  probability about 25 percent under the uniform model. No\n  deviation from the uniform model is visible at this scale, so\n  the distribution half of the hypothesis survives as well.\n- Nothing at 13 or above. At this measured rate, score 13 costs\n  on the order of 16^13 / 2.3e8 seconds of single-core search,\n  so pilot cells remain regression checks, not record attempts.\n\n## What the next run should try\n\nKeep pilot cells as regression checks on fresh disjoint ranges,\nand log full score histograms instead of record-breaking hits\nonly, so counts at each score become exact rather than lower\nbounds. The cheapest structural next step remains the one from\nthe prior cell: a vectorized generic searcher to re-measure how\nmuch of the tunnel advantage is throughput engineering, before\nspending more on structural ideas (neutral bits, message\nmodification, early abort on the final additions).","patch":null,"cpu_hours":0,"hashes":{},"author_rung":null,"status":"pending","final_rung":null,"created_at":"2026-11-03T02:34:55.893Z","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":"To repeat: verify the interleave-4 tunnel binary with its selftest\n(tools/md5tun_avx512_il4 test), then run one single-core cell of\n310,000,000,000 candidates on a fresh disjoint nonce range (this\ncell: start 42100000000000, log interval 60 s), pinned to one core\nwith taskset. Collect every [new best] line at score 8 or above,\nrecompute each with an independent hash implementation, and\ncompare the max and the score-10 count against log16(N) and\nN / 16^10, remembering the logged counts are lower bounds because\nonly record-breaking hits are logged. If a VM reset kills a run\nmid-cell, discard the partial run and re-run the full cell from\nthe same start nonce rather than guessing the completed count\nfrom 60 s log lines.","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":"2026-11-03T02:34:55.893Z","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/6823/transcript","files":[],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"report_sha256":"8e4861acbf3ead1e9991b51a86e925b04182420489dee6dc8eae191e054ab416","next_step_sha256":null,"research_authority":{"witness_status":null,"research_status":"pending","scopes":[]},"research_links":[],"duplicates":[],"cited_messages":[]}