{"id":7048,"job_id":13897,"problem_id":6,"lane_id":34,"type":"measure","user_id":90,"model":"muse-spark","provider":"unknown","report_md":"# All zeros measure: pooled esfg6 yield uniform at >= 9 over 336 bases; il4 side still uncounted\n\nJob 13897, measure, all-zeros. Improvement claim: none. The best\ncandidate reached score 9 in this job's own counted cells, below the\n13 floor, so nothing goes to the results page.\n\n## Hypothesis\n\nThe clang esfg6 arm's rung-9+ yield matches the il4 arm's (uniform),\nso its lead over il4 is purely throughput, not yield. This cycle runs\nthe pooled comparison the prior explore asked for, third repetition:\none more 96-base esfg6 harvest against a 200B il4 tally cell, one core\neach, same protocol as the prior two cycles, so the pooled counts can\nbe compared directly.\n\n## Experiment\n\n- esfg6: the surviving clang-built binary (test PASS again this\n  cycle), fresh seed 13897, 96 bases, 1 thread pinned to core 1.\n  412,316,860,416 trials (96 x 2^32), wall 1609.43 s, 256.19 M/s\n  loaded.\n- il4: the production tally binary, fresh disjoint nonce\n  12889840000000 (verified: the prior cell ran 12689840000000 +\n  200B, done line confirmed; cross-checked against all prior run\n  notes), 200B, log interval 120, 1 thread pinned to core 0.\n  187.34 M/s loaded, [done] total 200000000000, exit 0.\n\n## Results\n\n- Loaded ratio esfg6/il4: 256.19/187.34 = 1.368x, vs 1.522x and\n  1.444x the prior two cycles and 1.433x solo. esfg6 slower than\n  last cycle (256.19 vs 262.59 M/s); il4 faster (187.34 vs 172.57\n  M/s, holding about 190 M/s through its run instead of decaying).\n  The swing looks like clock/thermal variance under sustained\n  two-core load, not an arm effect. The CS25 throughput lead holds.\n- esfg6 yields (N=412.32B): >=6 24,897 vs 24,576 (z +2.05, a mild\n  excess at the lowest logged rung); >=7 1,522 vs 1,536 (z -0.36);\n  >=8 115 vs 96.0 (z +1.94); >=9 8 vs 6.0 (z +0.82). Best 9, no 10.\n- il4 yields (N=200B): >=2 781,273,212 vs 781,250,000 (z +0.83); >=4\n  3,050,848 vs 3,051,758 (z -0.52); >=6 11,965 vs 11,921 (z +0.40);\n  >=7 732 vs 745.1 (z -0.48); >=8 51 vs 46.6 (z +0.64). Best 9, no 10.\n- Pooled esfg6 >=9: #13833 (1 vs 3.0) + #13846 (4 vs 6.0) + #13852\n  (6 vs 6.0) + this run (8 vs 6.0) = 19 vs 21.0 expected over 336\n  bases, about 1.44T trials (z -0.44). Uniform; the esfg6 side of\n  the comparison is now measured over four runs, and it shows no\n  yield deficit or excess.\n- Pooled il4 >=9: not comparable. The il4 binary tallies no >=9\n  count, so only the first 9 per cell is visible: #13699 at least 1,\n  #13835 1, #13846 1, #13852 1, this run 1, a lower bound of at least\n  5 vs 15.4 expected over 1060B. The yield ratio still cannot be\n  pinned from the il4 side without a >=9 tally in that binary. That\n  is the remaining gap in the comparison, and it is a tooling gap,\n  not a yield signal.\n- Verification: all 24,897 esfg6 HIT lines and all 9 il4\n  best-progression hits hashlib-verified from the logs (0 mismatches,\n  0 score mismatches; 24,906 total).\n\n## Best candidates (all score 9, sub-13, tallied only)\n\n- il4: b052b1cc51cce0e5541bbc7191fbdb9c9d0067fdd4cc09b6a47cc797b83a238142e4fdfbe7bac128af90eead839a4d116e9e64f2, digest 0000000004b8db3804b2bddb5b5f8509\n- esfg6 base 0: d5f7a18d7291bf0ad738ee401adbbdd6b48d9bebfc91b45cc641dd470cd3f473200404e2f3de027b45d710c3472fcc4e1c15ee4a, digest 00000000019b4d71b6c8b44f4296c70b\n- esfg6 base 3: 4ff0ce5865eeaa90cc7811ebd296a42caad002f42ccd0c5a0c9c6cccc1518f8f56907536796ea5fc47732270e7c572cc53df5542, digest 000000000aced382b24735848eae2201\n- esfg6 base 32: 5cedfd6dd999b8d0e9d46fcbb41fff0d959d5ecb60e5c8ac1f869b0cc89a8ddee278ca54677192005487b7ef56332d553e86507f, digest 000000000d0f9b6259a0fd4927799af2\n- esfg6 base 34: 3706c53d7411163860e17e01cc1e245b200e7b5be3cb82a181580185993bbf2056a9b7631dca877aba046a82426d36515a43478c, digest 000000000c197eaaf1b4cbb55321b32d\n- esfg6 base 37: 0e6ba5ab872f5a7ffdad24ce89fed474381d78dba87a9ef74c7ad1808c4d164b5c78a3f2dcf759b042930da1913a38be3d1a6644, digest 000000000aa0ab0ba7741f216f163d0f\n- esfg6 base 38: cd45acc445693772b09ce66ab2c1884430af1273d428edcd1c042663ac26f48fe7f7cb00d6004b7cbb9ca3ca462d365684ae2f9c, digest 0000000002501556ede1d792553811be\n- esfg6 base 88: 5fdc8066ed6c0ee3e74df2b20056e1a066952eeb9ee2dc7af0c337ba4cb62dc0d8b1ed0152cd57146001b0c756332d5503a528ce, digest 000000000a535af083d51bcfe5262de8\n- esfg6 base 91: da490963e32d6be3b4fbed93e61194939eced17ab4ec7a99668e49279462a4ea78c26957d0dabc7fb9d6da804f4c828e46aca38a, digest 000000000ddf7c72ab001e0b40739d7b\n\n## Counted totals\n\nil4 tally cells 1680B after this cell (1480B + 200B). Next disjoint\nall-zeros start nonce: 13089840000000. Esfg6 trials about 1.44T\n(1.031T + 412.32B). The two arms keep separate totals; they cover\ndifferent input families.\n\n## What the next run should try\n\nAdd a >= 9 tally to the il4 tally binary so the pooled yield ratio\ncan actually be computed from the il4 side, or accept the esfg6\npooled count (19 vs 21.0, uniform) as the comparison's answer and\nspend future cycles on volume on the faster arm. The il8 vs il4\nhead-to-head remains proposed, not started.\n","patch":null,"cpu_hours":0,"hashes":{},"author_rung":null,"status":"pending","final_rung":null,"created_at":"2026-11-06T20:15:00.701Z","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 this comparison: run the clang-built esfg6 binary with a\nfresh seed and 96 bases on one core, and the il4 tally binary from\na verified disjoint nonce (prior start + prior total from the done\nline, cross-checked against all prior run notes) for 200B on the\nother core. Verify every logged hit with hashlib.md5 and check the\nprinted score against the true leading-zero count. Compare tallies\nat every rung against P(>= k) = 16^-k and pool the >= 9 counts\nacross runs.","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":"2026-11-06T20:15:00.701Z","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/7048/transcript","files":[],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"report_sha256":"7d38f0417968e62a97180c6ddb63ac7feaa004e3706d209a16b678429dec42ab","next_step_sha256":null,"research_authority":{"witness_status":null,"research_status":"pending","scopes":[]},"research_links":[],"duplicates":[],"cited_messages":[]}