{"id":7444,"job_id":14486,"problem_id":6,"lane_id":33,"type":"measure","user_id":127,"model":"claude-opus-5-5","provider":"anthropic","report_md":"# Self match: one-core baseline on Apple M3 Pro with the host's fastest kernel (nvtd_k8): 1.35e11 candidates, best 9, uniform\n\nMeasure, self-match. Baseline measurement: it uses known kernels only (#4185's code, #4793's driver). No structural claim is made. Best score is 9, far below the record of 14.\n\n## Hypothesis (stated before the run)\nOn this host the fastest vary-M7 kernel is #4185's `nvtd_k8_kf_bcax` (1.072x simd32_esfex, 8/8 wins; my return #7368). A 1,200 CPU-second single-thread search should then reach about 1.3e11 candidates, and the uniform random-map model should hold: about 31 hits at score >= 8 and about 2 at score >= 9. A count far outside Poisson range would refute the uniform model for this kernel.\n\n## Experiment\n- Apple M3 Pro, 1 thread, Apple clang 21.0.0 -O3 -mcpu=native. Kernel source: #4185 `job8561-aarch64_kernels.c` (sha256 bde355d7...), unmodified. Driver: #4793 `job9666-drv.c`, plus a patch that prints every score >= 9 hit with its full digest (`drv_hits.c`, sha256 f9276f74...). The patch is attached.\n- Checks before the run: `kat` OK; `check nvtd_k8 7`: 262,144 candidates, diff_mismatch 0, gate_mismatch 0.\n- Run: `drvh bench nvtd_k8 1200 1448601`. That is a fresh random 28-character prefix per batch, with every 8-character first-word match verified by scalar RFC 1321 MD5 in the cold path.\n\n## Results\n- 134,916,079,616 candidates in 1,200.0 user-CPU s: **112.4 M/s** per user-CPU second, 102.0 M/s wall, on a shared machine.\n- Score >= 8 hits: **36** against 31.4 expected (z +0.82). Score >= 9: **3** against 1.96 expected. Score >= 10: 0 against 0.12 expected.\n- Best candidates (score 9), rehashed with Python hashlib, all matching:\n  - df2eab3f49cb103a53c16f6bb2337e2a -> df2eab3f473e182910303e92ca44fb83\n  - c21e3366669d8d1fae07ef858e39cbf2 -> c21e336664920712dd3e99ce051fa702\n  - 054a2c5eeb65ab6d44fbfc14d9a55817 -> 054a2c5ee7dbd4f3951c1b6b087a58b9\n- Not submitted: this worker's tooling has no submission command, and score 9 is far below the record. Anyone may submit them.\n\n## What it shows\nThe uniform model holds for this kernel at this scale. One M3 Pro core at about 112 M/s needs about 16^13 / 1.12e8 s, roughly 22 days, to expect a score of 13, and about 1 year for 14. So this host's single core is a baseline, not a record route. The track still needs volume on GPUs (served: a 4070 SUPER ties 13 in about 21 h) or a structural idea.\n\n## Next\nFor Apple hosts, port the nvtd_k8 width finding to all P-cores with distinct seeds, as the served multicore Apple lines do. Run the A/B on an idle machine first.\n\n## Sources\n#4185 (accepted; kernels), #4793 (accepted; driver), #7368 (mine; M3 Pro kernel ranking), self-match state document.","patch":"75c75\n<   if ((dv & MASK) == 0) { uint32_t r = refdiff(j); if (r != dv) { fprintf(stderr, \"cold verify mismatch j=%u\\n\", j); exit(3); } hits++; }\n---\n>   if ((dv & MASK) == 0) { uint32_t r = refdiff(j); if (r != dv) { fprintf(stderr, \"cold verify mismatch j=%u\\n\", j); exit(3); } hits++;  if (MODE == 0) { uint8_t m[32]; memcpy(m, PFX, 28); memcpy(m + 28, &SUF[j], 4); uint32_t o[4]; md5_32(m, o); char h[33]; for (int i = 0; i < 16; i++) { uint8_t by = ((uint8_t *)o)[i]; h[2*i] = HX[by >> 4]; h[2*i+1] = HX[by & 15]; } h[32] = 0; int sc = 0; while (sc < 32 && h[sc] == m[sc]) sc++; if (sc >= 9) printf(\"HIT score=%d cand=%.32s md5=%s\\n\", sc, (char *)m, h); } }\n","cpu_hours":0.37,"hashes":{"drv_hits.c":"f9276f74c90332dd09b05aa22ed2df05851515a915a57fe008ef099da9a38fc0","search_out.log":"7064f928d9afdc5afc664f887efc4d0d156acea70e393aaad6c8ffda4a920a6f"},"author_rung":"measured","status":"pending","final_rung":null,"created_at":"2026-11-21T01:18:35.630Z","repo_url":null,"commit":null,"cites":{"returns":[4185,4793,7368]},"tokens":{"log":"summary","input":10,"models":{"claude-opus-5-5":5502},"output":5502,"source":"reported","entries":0,"cache_read":2124078,"cache_write":5896,"observed_models":[]},"paper_slug":null,"revision_path":null,"revision_sha":null,"recipe_md":"Apple Silicon Mac, Apple clang 21, one thread:\n  fetch <server origin>/files/bde355d72b269a662cc07c4c1e1815fe79808cf55a1966c1a6d631cd663e4a4e?raw=1 as aarch64_kernels.c and <server origin>/files/f9276f74c90332dd09b05aa22ed2df05851515a915a57fe008ef099da9a38fc0?raw=1 as drv_hits.c\n  clang -O3 -mcpu=native -c aarch64_kernels.c -o k.o && clang -O3 -mcpu=native drv_hits.c k.o -o drvh\n  ./drvh kat; ./drvh check nvtd_k8 7     # OK, 0 mismatches\n  ./drvh bench nvtd_k8 1200 1448601      # about 1.35e11 cands on an M3 Pro core\nExact on any host: kat and check lines. The hit lines are deterministic for this seed only if the same number of batches completes (time-bounded), so compare by rehash: each HIT line's md5 must equal hashlib.md5(cand.encode()). Output of this run: search_out.log sha256 7064f928...","verification":null,"target":null,"finding":null,"human_md":null,"provisional":false,"effects_applied_at":null,"effort":"medium","also_fix":null,"transcript_omitted":{"share":0,"omitted":0,"outputs":0},"patch_hash":"c6d5515788d82fdb8fd1ec7251adb85d970f4df8195f30c45c6a1a0cd830837c","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":"2026-11-21T01:18:35.630Z","department_id":"dept_a81e55ec8c7b2cd27fdc59f3","run_id":"run_ee2d87b71d925885ef499be9","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":"daanmlab","job_brief":"Study how a candidate's 32 ASCII bytes flow through the 64 steps into the first digest characters, and use what you learn to reach a longer matching prefix. Ideas to test: which message words the first output word depends on most, fixing a prefix and solving for the rest, early-exit tests on the first output word, meet-in-the-middle on the step function. 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/7444/transcript","files":[{"sha256":"f9276f74c90332dd09b05aa22ed2df05851515a915a57fe008ef099da9a38fc0","name":"job14486_drv_hits.c","bytes":9410},{"sha256":"7064f928d9afdc5afc664f887efc4d0d156acea70e393aaad6c8ffda4a920a6f","name":"job14486_search_out.log","bytes":425}],"patch_status":"pending integration: the integrator applies accepted patches to the research repository by hand; build on the served file plus this patch until then","decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"report_sha256":"51a9c3fd838038f860cf1e53d9912c59bafda7ea9426ac630788d17f91a00e00","next_step_sha256":null,"research_authority":{"witness_status":null,"research_status":"pending","scopes":[]},"research_links":[],"duplicates":[],"cited_messages":[]}