{"id":6814,"job_id":13547,"problem_id":6,"lane_id":33,"type":"explore","user_id":90,"model":"muse-spark","provider":"unknown","report_md":"# Self-match study: the nvtd arm still holds about 316 MH/s on this VM, about 7x the local binary\n\nJob 13547, explore, self-match lane. Improvement claim: none. No\ncandidate scoring 13 or above exists, so nothing goes to the\nresults page. The best hit in this cycle scored 9 and is recorded\nhere and in the tally only.\n\n## The uncovered obligation\n\nThe previous explore cycle measured the local AVX-512 vary-M7\nbinary at 41.54 M/s on this VM, but the nvtd arm's 305.2 MH/s\nnumber dated from October, before several VM resets. A planner\nchoosing a campaign arm on the current instance therefore had\none fresh number and one stale one. This cycle closes that gap:\nit re-benches the nvtd arm on the current instance, with the\nsame binary and source hashes as the October campaign, and runs\na search cell with it.\n\n## Accepted scoped answers compared\n\nThe fixture scores 12. The October campaigns reached score 11,\nwith hit tallies consistent with the uniform 16^-k null, and\nmeasured a two-worker colocation penalty of about 0.68x on this\nVM. The local binary's 41.54 M/s from the previous cycle stands.\nNone of that changes here. The open question this cycle closes\nis narrower: does the nvtd arm still run near 300 MH/s on the\ncurrent VM instance, and what does a search cell with it yield.\n\n## Experiment\n\nBinary identity verified before the run: runs/selfmatch-20261019/sm_pgo\nsha256 2ce4d4a0bfefdcaeff45a2c04b96c76bc0c14a74bb3ab966cfd53e279fb64105,\nsource selfmatch_nvtd_x86.c sha256\n6eef18a32be88476938b6aa0bde3863c3d22b0596d58ee5ef2b17dd2e0469ae2,\nmatching the October campaign. Selftest passed. Bench, single\nthread, arm simd96_nvtd_padtl_kf, seed 99991, 30000 batches,\nthree repeats: 317.451, 316.978, 315.658 MH/s, mean 316.7 MH/s,\nbest 8 in all three, same best candidate each time. The track\nfixture was recomputed with the system md5sum binary: score 12,\nas in the brief.\n\nSearch cell: single thread, arm simd96_nvtd_padtl_kf, seed\n13547000000001, 1800 seconds, minimum logged score 9. It ran\npartly in parallel with a measure cell on the second core, then\nalone after that cell finished. Results: 510,895,587,328\ncandidates, 283.83 M/s average over the 1800 seconds, best\nscore 9. Five hits at score 9, none at 10 or above. Expected\ncounts under the uniform model: 7.43 at 9 or above, 0.46 at 10\nor above, so 5 and 0 are both in line. All five hits were\nrecomputed with md5sum and agree at score 9:\na8949145b8f93b77797d0206b8f1ab02,\na0ea72fbe463c5f5cac3496369519443,\n7b72e3936a2e5dee0a2d6dd29e079a26,\na5e734e1d44ba9b265b32489b69a70f2,\n0bb13f1af264bc515cc19a009d0d7aa4.\n\n## Changed premise\n\nThe nvtd arm still holds on the current instance: 316.7 MH/s\nsolo bench, against 305.2 MH/s in October and 312.7 MH/s in the\nearlier bench on this VM. That is 7.6x the local binary's\n41.54 M/s, so the previous cycle's changed premise stands and\nis now confirmed on fresh numbers rather than a stale one:\ncampaign planning should use the nvtd build for any large\nself-match cell, not the local binary. The search average of\n283.83 M/s, below the solo bench, is the expected two-worker\nload effect for the first part of the cell, not a regression.\nSizing at the solo bench rate: a score-10 hit needs about\n1.1e12 candidates, roughly 58 minutes single-core; at the\nlocal binary's rate the same expectation is about 7.4 hours.\n\n## Cheapest next experiment\n\nA two-core nvtd search cell with disjoint seeds, sized to\nabout 2e12 candidates, would bring the expected score-10 count\nto about 1.8 and test the null at the next rung on the\ncampaign arm itself, rather than on the slower local binary.\nThat cell needs about 53 minutes at the solo bench rate on\none core, or about half that split across two cores, so it\nfits a later cycle's window but not this cycle's remaining\nwindow, and is proposed, not run.\n","patch":null,"cpu_hours":0,"hashes":{},"author_rung":null,"status":"recorded","final_rung":"recorded","created_at":"2026-11-02T22:31:18.062Z","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 runs/selfmatch-20261019/sm_pgo by sha256\n(2ce4d4a0bfefdcaeff45a2c04b96c76bc0c14a74bb3ab966cfd53e279fb64105),\nrun its selftest, then run bench simd96_nvtd_padtl_kf 99991\n30000 1 three times and save the output as bench.txt. Expect\nabout 316 to 317 MH/s and best 8 with candidate\n8dddf22b227e31e673b6962c63140db0. Then run search\nsimd96_nvtd_padtl_kf 13547000000001 1800 1 9 and save the\noutput as cell.log. Expect about 5.11e11 candidates, about\n284 M/s average, best 9, five score-9 hits. Verify with\nprintf '%s' 'a8949145b8f93b77797d0206b8f1ab02' | md5sum, which\nshould give a8949145bfa7e3a9a5ac94ff40fe9d75, and verify the\nfixture the same way for score 12. Compare the bench mean\nagainst 305.2 MH/s for the October nvtd number and 41.54 M/s\nfor the local binary on this VM.","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":null,"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":"Identify an uncovered obligation or a changed premise on this track; compare the accepted scoped answers before proposing the cheapest new experiment. Deliberate replication needs a stated independence objective.","review_deferred":false,"in_triage":false,"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/6814/transcript","files":[],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"report_sha256":"d513d76a7afc1ffb34dc5ffd8549050243783319bb3ee5fe4e2aa6a3a1f287d8","next_step_sha256":null,"research_authority":{"witness_status":null,"research_status":"recorded","scopes":[]},"research_links":[],"duplicates":[],"cited_messages":[]}