{"id":6993,"job_id":13783,"problem_id":6,"lane_id":35,"type":"measure","user_id":90,"model":"muse-spark","provider":"unknown","report_md":"# Smallest collision: longer pilots on ca5532 and f10020 do not reach Q31, and run slower than the 100 s pilots\n\nJob 13783, measure, smallest-collision. Improvement claim: none.\nNo full collision pair was produced, so nothing goes to the results\npage. The hypothesis under test, from job 13777's cheapest next\nexperiment, was that longer pilots on f10020 and ca5532 would push\nbest depth past Q30 toward the Q31 the pair-cost projections assume.\nIt did not: at 300 s per seed the best depth is still 30, with no\nQ31 on either arm, and the measured rates fall to about 0.55-0.61x\nthe 100 s pilot rates on the same machine.\n\n## Experiment\n\nBuilt the two gxr_fast arms from the published job9002-gxr_fast.c\n(gcc -O3 -march=znver4 -mtune=znver4 -DFIXQ12): ca5532 at -DPADL=63\n(L=63, T9=27), f10020 at -DPADL=62 (L=62, T9=25). Pilots: 3 fresh\nseeds (13783-13785) x 300 s x 1 thread per arm, pinned to core 0\n(a parallel smallest-collision explore job ran on core 1). The\nfirst attempt at these pilots was killed mid-run by a VM reset;\nthe run was relaunched cleanly from the same plan with the same\nseeds, and only the completed relaunch is reported here. The\nkilled partial logs are preserved in the run dir and contribute\nnothing to the numbers below.\n\n## Results\n\nQ27 rates, 900 thread-s pooled per arm (mean of 3 seeds, sd\nacross seeds):\n\n| arm | L | T9 | ntup | Q27 | Q27/thread-s | vs served (13777) | pair log2 s (H) |\n|---|---|---|---|---|---|---|---|\n| ca5532 | 63 | 27 | 1024 | 1,032,679 | 1145.70 +/- 12.47 | 12.36x | 22.44 |\n| f10020 | 62 | 25 | 256 | 271,954 | 301.67 +/- 35.14 | 3.25x | 24.36 |\n\nDepth reached (Q27/Q28/Q29/Q30/Q31 totals): ca5532 1,032,679 /\n3,250 / 248 / 15 / 0; f10020 271,954 / 924 / 35 / 1 / 0. Best\ndepth 30 on both arms. No Q31. All 4,164 emitted records (6\nbest-per-seed plus every d=28/29 log record) verified\nindependently with verify_best_13783.py (depth against the path\nheader, padding bytes, hashlib digests, T9 tunnel hold), 0 bad.\n\nTwo findings:\n\n1. The Q31 hypothesis is refuted at this length. Tripling the\n   pilot length from 100 s to 300 s per seed moved neither arm\n   past depth 30: ca5532 produced 15 Q30 records and f10020 one,\n   and zero Q31 in 1.30M Q27 attempts. The pair-cost projections\n   that assume a Q31 tail are still unmeasured, and longer\n   single-seed runs are not the way to reach it.\n2. Longer pilots run slower, not just longer. ca5532 measured\n   1145.70 Q27/thread-s against 1891.82 in the 100 s pilots of\n   job 13777 (0.61x); f10020 measured 301.67 against 546.62\n   (0.55x). The per-seed rates are stable within each run\n   (ca5532 sd 1.1% of the mean), so this is a level shift between\n   the two run lengths, not seed noise. The process CPU time ran\n   at 0.66-0.72x wall time in the 300 s runs, so the machine gave\n   each single-threaded pilot less than a full core under\n   sustained two-core load; whether the shortfall is host\n   throttling or a warmup burst in the first 100 s is not\n   separated by this experiment. Rates should be quoted with\n   their run length until that is separated.\n\n## Answer\n\nLonger pilots do not buy depth: best depth stays 30 and no Q31\nappears at 300 s per seed, so the Q31 tail the 124B/126B pair\ncosts assume remains unmeasured. The 126B pair via ca5532 still\nprojects at 2^22.44 thread-s locally (H) and the 124B pair via\nf10020 at 2^24.36, both below the served 128B cost, but the\nrates behind those projections are length-dependent: the 300 s\nnumbers are roughly half the 100 s numbers. Any pair-cost claim\nshould state its pilot length.\n\n## Cheapest next experiment\n\nRun one arm at 2 threads (both cores, 300 s) against its\n1-thread 300 s rate: if the per-thread rate recovers toward the\n100 s number, the shortfall is thread/host contention; if it\nstays near 1145/s, the 100 s number was a warmup burst. In\nparallel, a connect run on the ca5532 survivors would test\nwhether depth 30 pairs connect at all, which is the cheaper\nquestion than chasing Q31 by brute pilot length.\n","patch":null,"cpu_hours":0,"hashes":{},"author_rung":null,"status":"pending","final_rung":null,"created_at":"2026-11-05T21:07:27.892Z","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. Build gxr_fast arms from job9002-gxr_fast.c with gcc -O3\n   -march=znver4 -mtune=znver4 -DCONDS='\"<header>\"' -DFIXQ12:\n   ca5532 with -DPADL=63, f10020 with -DPADL=62.\n2. Smoke-run each binary for 2 s and check the JSON line carries\n   the published T9 width (27/25).\n3. Run each arm 3 fresh seeds x 300 s x 1 thread\n   (./gxr_<arm> <seed> 1 300 28), saving the JSON summary line per\n   run. Pool Q27 over the 900 thread-s per arm.\n4. Verify with verify_best_13783.py (best record per seed plus\n   every d=28/29 log record, independent integer steps): expect\n   0 bad.","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":"2026-11-05T21:07:27.892Z","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 how MD5 collisions are built (differential paths, message modification, the single-block attacks of Xie and Feng and Stevens) and what limits their length, and use it to find a shorter full collision. Running fastcoll gives 128 + 128 bytes from known techniques; it is the baseline to measure against. Ideas to test: where the single-block attacks spend their work, whether a shorter second member or a shared prefix can change the bound, what a 64 + 64 search would cost on your machine. 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/6993/transcript","files":[],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"report_sha256":"519d68c8c5006e2e37a27b8ecfa03b8826ca8e7e4e1c12cb2dacb7685e966df7","next_step_sha256":null,"research_authority":{"witness_status":null,"research_status":"pending","scopes":[]},"research_links":[],"duplicates":[],"cited_messages":[]}