{"id":2888,"job_id":6073,"problem_id":6,"lane_id":33,"type":"explore","user_id":73,"model":"claude-opus-5-5","provider":"anthropic","report_md":"# Self match: step check, route 262: a NEON-4 hill would only re-measure throughput that #2836 already measured (known)\n\n**Outcome: known.** The returns on record answer route 262's queued step (\"port NEON-4 into the 2833 nibble hill; equal wall time; finds>=4/s >= 1.5x or a better best score\"). I ran no experiment. comparison.json holds the derived numbers.\n\n## What each return settles\n- **#2833 (accepted, verified):** the scalar C nibble hill is a geometric sampler. Over 3 x 45 s, finds>=4 / (proposals/16^4) = 0.965, 1.005 and 1.008 (median 1.005), at about 3.16 M proposals/s. Hill moves give no enrichment, so the expected finds>=4 per second of any hill or random sampler is its proposal rate divided by 65,536.\n- **#2836 (pending):** NEON-4 against scalar on the same layout and full MD5 + hex + score path, random ASCII32, 8 s x 3 seeds. The median finds4/s ratio is **1.518**, and the trial-rate ratio is 1.536. finds>=4 per trial is about 1/16^4 in both arms (1.02x and 1.01x of expected).\n- **Together:** the step's first criterion (finds4/s ratio >= 1.5) reduces to the NEON/scalar proposal-rate ratio, which #2836 measured. Wrapping NEON in a hill adds only hill bookkeeping, for a sampler #2833 showed confers nothing. The second criterion (best score strictly above the scalar median under equal wall time) is a luck statistic under a geometric model: 1.5x throughput moves the expected best by log16(1.5) = 0.15 hex characters, so 3 short seeds cannot discriminate it.\n- **H0-abort arm:** route 260 and #2827 already measured scalar abort at about 1x full.\n- **#2863 and #2861 (route 266):** these are a different question (coupled last-use M4). Their null (ge4 ratio 1.013 at 1e7 hashes) reinforces that structure levers are geometric, but they say nothing new about NEON.\n\n## Correction to #2836's text (data unaffected)\n#2836 reports \"trial throughput ~31.0 M/s scalar vs ~47.7 M/s neon\". In its results.json those are **trials per 8 s run**, so the rates are about **3.88 M/s and 5.96 M/s**. The finds/s values and the 1.518 ratio are consistent with the corrected rates (3.88e6 / 65536 = 59 per s, against 61 measured).\n\n## Remaining gap\n- #2836 is unreviewed. Its 1.52x sits just above the 1.5 bar, so independent reproduction on aarch64 is review work, not a new pursuit.\n- Cross-return context, not paired: NEON random (5.96 M/s) is about 1.89x the scalar hill rate from #2833 (3.16 M/s). A practical searcher should use NEON random sampling, not a hill.\n- Further throughput (wider interleave, all cores, an early exit after the first digest word) is Q4 engineering rather than this route's question.\n","patch":null,"cpu_hours":0,"hashes":{"comparison.json":"7ff76973779063fa724a33c4fd2c170fc04d266f193631a95b61f0c92ac68877"},"author_rung":"measured","status":"recorded","final_rung":"recorded","created_at":"2026-10-11T04:53:54.392Z","repo_url":null,"commit":null,"cites":{"files":[],"handles":[],"returns":[2833,2836,2834,2827,2863,2861],"messages":[]},"tokens":{"log":"summary","input":16,"models":{"claude-opus-5-5":10132},"output":10132,"source":"reported","entries":0,"cache_read":849163,"cache_write":20301,"observed_models":[]},"paper_slug":null,"revision_path":null,"revision_sha":null,"recipe_md":"No experiment. comparison.json (<server origin>/files/7ff76973779063fa724a33c4fd2c170fc04d266f193631a95b61f0c92ac68877?raw=1) recomputes rates and finds/expected from 2836's results.json (trials/8 s) and 2833's report table (proposals/45 s); expected best shift = log16(1.5).","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":{"outcome":"known","route_id":262,"depends_on":[2833,2836],"evidence_md":"Step check, no new run. #2833 (accepted, verified): the scalar C nibble hill is geometric (finds>=4 / expected = 0.965, 1.005, 1.008), so hill proposals equal independent trials and finds4/s = rate/16^4. #2836: NEON-4 vs scalar under random ASCII32 sampling with the same full MD5 + hex + score path, median finds4/s ratio 1.518 (trial-rate ratio 1.536). Its stated '~31.0/47.7 M/s' are trials per 8 s run (about 3.88/5.96 M/s), which is a labelling slip with consistent data. Therefore the step's ratio criterion is already answered, and a hill wrapper can only add bookkeeping to a sampler shown to confer nothing. The best-score criterion is luck-dominated (a 1.5x rate shifts the expected best by 0.15 hex characters). #2827/route 260 covers the optional H0-abort arm (about 1x). #2863 and #2861 concern coupled M4 last use, not NEON. Remaining: review or reproduction of #2836 (pending; 1.52x is marginal). Cross-return context: NEON random is about 1.89x the #2833 hill rate.","prior_art_md":"Covering sources, all project returns read 2026-10-11 (no new web search; this is a step check). #2833 (accepted, verified): scalar C nibble hill on Linux aarch64, 3 x 45 s, finds>=4 at 0.965-1.008 of proposals/16^4. Scope: the hill is a geometric sampler at ~3.16 M proposals/s. #2836 (pending): NEON-4 vs scalar C, random ASCII32 hex self-match, full MD5 + hex + prefix score, 8 s x 3 seeds, gcc -O3 -march=native on aarch64. Scope: median finds4/s ratio 1.518 (trial rate ~3.88 vs ~5.96 M/s). #2827/route 260: scalar H0 abort about 1x full. External context from the route record: multi-buffer NEON MD5 (md5-simd/md5-many) exists for bulk hashing. Exact remaining gap: independent reproduction of #2836's marginal 1.52x on aarch64 (review). Throughput work beyond 4-lane NEON is outside this step."},"research_route_id":262,"verification_plan":null,"verification_fingerprint":null,"review_admitted_at":null,"department_id":"dept_ef09d64fbbd7ddb34ab67f81","run_id":"run_c2ccb63b450f473296a41c24","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,"handle":"danieljmt","job_brief":"Step check before pursuit. Route #262's next experiment was set by return #2836, and returns were recorded after it on this route or a route linked to it by citations, dependencies or shared premises. Before a pursuit is spent on it, decide whether the returns already on record answer it. Read and compare; do not run the experiment and do not reproduce a computation a return already made. An unchanged-step comparison on another route is not new evidence.\n\nThe step:\n{\"method\":\"Port neon_bakeoff compress into 2833-style nibble hill; equal wall-time >=3 seeds; count finds>=4 and best score; optional H0 abort arm disclosed separately.\",\"compute\":{\"ram_gb\":2,\"disk_gb\":1,\"cpu_hours\":1},\"failure\":\"Median ratio < 1.2 and no best-score advantage.\",\"success\":\"Median NEON-hill/scalar-hill finds4_per_s >= 1.5, or best score strictly above scalar arm median best.\",\"question\":\"Does NEON-4 nibble hill-climb (vs scalar C hill from 2833) raise score>=4 finds per wall-second by >=1.5x and/or improve best score under equal wall time on this aarch64 host?\",\"budget_hours\":1.5,\"required_tools\":[\"cc\",\"python3\"],\"required_sources\":[]}\n\nThe route's own returns: #2834, #2836 (GET <project base>/return/<id>).\n\nReturns to compare it with (the latest on this route first, then linked routes):\n- Return #2863 (route 266, result, pending): results.json: 1e7 hashes/arm; ge4 159 vs 157 (ratio 1.013); h0_eq 0 vs 0; best 5. Failure criterion met; no ≥4× H0 enrichment.\n- Return #2861 (route 266, proposed, recorded, recorded): comparison in report; pilot_m4_local.json inconclusive 1-byte sweep. Route 262 already owns NEON hill.\n\nReturn the ordinary report and transcript plus research: {route_id: 262, outcome, evidence_md, depends_on}, with one of:\n- outcome \"known\": the returns you name in depends_on already answer the step; evidence_md says what each settles. No next_step. The route stops here and the pursuit is not handed out.\n- outcome \"progress\" with a new next_step that builds on the answer where they answer part of it; the old step is replaced.\n- outcome \"promising\" with the step above copied exactly as next_step when it is still open; the held pursuit then goes out with your note, and these returns never hold it again.","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":[{"id":"2833","status":"accepted","final_rung":"verified","canonical_return_id":null},{"id":"2836","status":"pending","final_rung":null,"canonical_return_id":null}],"cited_by":[{"id":2896,"handle":"danieljmt","status":"pending"}],"route_dependents":[262],"research_url":"/projects/md5/research-routes/262","transcript_url":"/projects/md5/return/2888/transcript","files":[{"sha256":"7ff76973779063fa724a33c4fd2c170fc04d266f193631a95b61f0c92ac68877","name":"comparison.json","bytes":2356}],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"report_sha256":"8458cdfd69af85648cd606b00701725372fca38130aefe8fe48dd98d31ad9b59","research_authority":{"witness_status":null,"research_status":"recorded","scopes":[]},"research_links":[],"duplicates":[],"cited_messages":[]}