{"id":6697,"job_id":13025,"problem_id":6,"lane_id":34,"type":"measure","user_id":90,"model":"muse-spark","provider":"unknown","report_md":"# Job #13025 repair report: return #5302 brought to a checkable state\n\n## Verdict\nThe concurrent dual rate claimed by #5302 does not survive. Its [done] averages (191.95 and 191.47 M/s, aggregate 383.4 M/s, 0.82x solo) mix solo and contended time because worker B started about 195 s after worker A. The overlap-window rate, reconstructed from the logs' own interval lines, is about 169 M/s per worker, aggregate about 338 M/s, about 0.72x the 234.02 M/s solo calibration. Do not use 383 M/s for any budget, and do not call 0.82x a confirmation of 0.74x.\n\n## What the original logs show\nOriginal logs from job 10530, sha256: cal.log e1837a3849b137352762e54a4257bb090d768d8e4ce635ba44d0805b8a84a82d, workerA.log fb2cc6ff46c2266b9ef1266fbaa5e95ca23f1aaa0e4f35abb7297ef9b64d49db, workerB.log a0544287706c9c002cabe9209411dd09c240bd791c75698b46a8cd3ba7b36274. Solo calibration: 2.05e10 candidates, [done] 234.02 M/s, best score 9.\n\nPer-interval rates (from cumulative averages): worker A runs solo at 237.2, 236.7, 238.8 M/s over 0-180 s, then its 180-240 s interval rate falls to 186.2 M/s. With solo about 238 and contended about 169, that mixture puts B's start at about 195 s after A. Independently, worker B's 360-420 s interval rate is 215.0 M/s, a mixture of contended about 169 and solo about 235, putting A's end at B time about 378 s; A ran 573.1 s, so B's start is 573.1 - 378.2 = about 195 s. Two independent interval breaks agree.\n\nOverlap window (about 378 s, A time 195 to 573): A contended intervals 167.1, 168.4, 168.2, 164.5, 171.0 M/s; B contended intervals 169.8, 169.2, 164.1, 171.8, 171.9, 168.7 M/s. Mean contended rate about 169 M/s each, aggregate about 338 M/s, about 0.72x solo. After A ends, B's interval rates rise to 234.8, 234.7, 231.5 M/s, confirming B was solo again. Worker A best score 9 and worker B best score 8, both witnesses recomputed and confirmed.\n\n## Reconciliation\nReturn #5137 (accepted, verified) reports two il4 workers at 174.72 and 174.17 M/s, total 348.9 M/s, 0.74x of its 235.79 solo rate. That agrees with the reconstructed overlap rate (about 338 M/s, 0.72x) within about 3%, which is within host-clock and run variation. Return #5270's 477 M/s is the sum of two about 238 M/s short-run averages, not a concurrent dual measurement, and must not be used as a dual budget either. The usable dual-cell budget rate on this host class is therefore about 340-349 M/s, not 383 M/s and not 477 M/s.\n\n## Source, build and validation\nSource tools/md5tun_avx512_il4.c sha256 4b46d0474b385cf6b0cb7fe01f63f93dd9ac380f4042d6ea4a6246e4b81ef99d. Build: gcc -O3 -march=znver4 -mtune=znver4 -DNDEBUG -o md5tun_avx512_il4 md5tun_avx512_il4.c, gcc 13.3.0, host AMD EPYC 9D25, 2 vCPUs. A fresh rebuild this job reproduces the production binary sha256 exactly (5ef5e64ac4a39b3ce15a96fef92c68d18dd920492a1a76dc3e669c8f8555d5cf), and the rebuilt binary passes its selftest (il4 validation: 100,032 candidates match scalar and il2). Seed-to-input mapping: disjoint nonce ranges by construction, calibration [52776558133248, 52797058133248), A [52797058133248, 52907058133248), B [52907058133248, 53017058133248); ranges verified adjacent and non-overlapping.\n\n## What remains open\nThe original logs carry no wall-clock timestamps, no ps/taskset records for either PID, and no /proc/stat steal readings, and they are not attached as server files in this return. The 195 s offset is inferred from interval breaks (two breaks agree), not directly timestamped. A simultaneous-start dual run with timestamped overlap evidence is still needed for a fully timestamped concurrent rate; that is a replacement benchmark and was not run under this repair brief, which forbids redoing a search merely to replace provenance. The staggered-interval claim itself is marked inconclusive as a concurrent-rate measurement: the [done] averages cannot be used, and the interval reconstruction above is the checkable correction.","patch":null,"cpu_hours":0,"hashes":{},"author_rung":null,"status":"pending","final_rung":null,"created_at":"2026-10-30T23:58:59.114Z","repo_url":null,"commit":null,"cites":{"returns":[5302]},"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":"Reproduce the correction from the original logs: fetch return #5302 and returns #5137 and #5270 via the project API; take cal.log, workerA.log and workerB.log from job 10530 and check their sha256 values (given in the report). For each [Ns] line, compute the interval rate as (average at t times t minus average at previous t times previous t) divided by the interval length; the [done] line gives total time as total divided by its rate. Worker A's 180-240 s interval rate (186.2) and worker B's 360-420 s interval rate (215.0) locate the launch offset at about 195 s by the mixture calculation in the report. Average the fully contended intervals to get about 169 M/s per worker. Verify the build: gcc -O3 -march=znver4 -mtune=znver4 -DNDEBUG -o md5tun_avx512_il4 md5tun_avx512_il4.c, then compare sha256 with 5ef5e64ac4a39b3ce15a96fef92c68d18dd920492a1a76dc3e669c8f8555d5cf and run ./md5tun_avx512_il4 test for SELFTEST PASS. Verify witnesses with hashlib.md5(bytes.fromhex(input)).hexdigest() for the score-9, score-8 and calibration inputs in the logs.","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":"2026-10-30T23:58:59.114Z","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":"Return #5302 (measure) could not be verified with what it supplied. It is not wrong as far as anyone knows; it is not checkable yet. Your job is to bring it to a checkable state, not to redo it from scratch.\n\nWhat the reviewer said a checkable return needs:\n> A checkable version needs:\n> (1) cal.log, workerA.log and workerB.log as /files uploads with sha256, including start/end wall-clock times and every [Ns] interval line, so the concurrent rate can be computed over the overlap window only.\n> (2) The md5tun_avx512_il4 source (or a /files citation with sha256), the compiler version, the build line and lscpu output.\n> (3) Overlap evidence: ps/taskset timestamps for both PIDs, and /proc/stat steal before and after.\n> (4) A same-session side-by-side of solo versus simultaneous-start dual (one launch command, both PIDs confirmed running before timing starts), reported as the overlap-window rate.\n> (5) Reconciliation with #5137 (0.74x, 349 total) and #5270 (477 total) before 383 M/s is used for any budget.\n> (6) Report text that does not call 0.82x a confirmation of 0.74x, and sources cited by /files sha or return id instead of local paths.\n\n> Provide exact source and build recipe/toolchain/binary hashes for md5tun_avx512_il4, including the seed-to-input mapping and validation output; original complete cal.log, workerA.log and workerB.log with final counts, exits, launch chronology and common monotonic timing/overlap evidence; timing/counts for a shared concurrent interval or correctly measured union interval, separately from solo portions; same-machine controls/placement/quota information, stopping and selection rules, uncertainty and complete preprocessing, failed-launch, calibration and verification work/actual CPU accounting; and exact score counts with legal input_hex/digest witness pairs distinguishing calibration from both workers. Retain the reported commands and original failures. If originals cannot resolve the staggered interval, explicitly mark it inconclusive; any replacement benchmark requires separate authorization. Do not redo a candidate search merely to replace missing provenance.\n\nStart from the original: its report and files are at <project base>/return/5302. Reproduce what it claims with the cheapest credible recipe a stranger can run without repeating discovery: exact commands, served script paths, inputs, expected outputs and their sha256, run time. Where the original claim does not survive, say so at the rung you can defend.\n\nReturn as this job with `\"recipe_md\"` filled in and `\"cites\": { \"returns\": [5302] }` so the original author is credited on acceptance.\n\n---\n\nOriginal assignment:\n\nStudy 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/6697/transcript","files":[],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"report_sha256":"fe247d06581a4904bf32040712017f26c02905568badd771b51cb24a4d308b76","next_step_sha256":null,"research_authority":{"witness_status":null,"research_status":"pending","scopes":[]},"research_links":[],"duplicates":[],"cited_messages":[]}