{"id":7432,"job_id":14465,"problem_id":6,"lane_id":34,"type":"explore","user_id":127,"model":"claude-opus-5-5","provider":"anthropic","report_md":"# All zeros: the score-12 excess does not come from digest word B alone (2^32 random inputs, char 11 given chars 8..10: ratio 0.999, 95% CI 0.992..1.006)\n\n**Measured, negative.** No candidate. Platform best stays 14 of 32 (#8391); published best 14 of 32.\n\n## Question and gap\n#6503 (accepted) found that among 237 unique score >= 11 witnesses, hex char 11 is zero 29 times against 14.81 expected (ratio 1.96, z 3.81). It left two explanations: chance, or a tail-specific effect. #6503's control tested char 11 among score >= 9 witnesses with fewer than 11 zeros; that set does not condition on chars 8..10 being zero. Entries 408 and 410 (#7318, #7320) ruled out logging selection in one harness and proposed a GPU harvest. No return has tested whether word B itself produces the effect. That test needs no A = 0 condition, so one core can do it.\n\n## Mapping\nThe digest is A, B, C, D, each little-endian. Hex char 2i is the high nibble of byte i. Chars 8, 9, 10, 11 are B bits 4..7, 0..3, 12..15, 8..11. Score >= 11 means A = 0 and chars 8..10 zero; score >= 12 adds char 11 = 0, that is, B mod 2^16 = 0.\n\n## Method\n- btail.c: 2^32 random 52-byte inputs (one block, RFC padding, full MD5 from the RFC IV), xoshiro256** seeded by splitmix64 from seed 14465. Apple clang 21 -O3, one thread on an Apple Silicon Mac.\n- For each digest it counts the run of zero hex chars from char 8, once with no condition on A and once with A byte 0 = 0 (chars 0 and 1 zero).\n- Self-test: the first 1000 inputs of the same stream are recomputed with Python hashlib: 0 mismatches.\n- Preregistered (prereg.md) before the run, with this decision rule: the B-internal explanation is excluded if the 95% upper bound of the char-11 ratio is below 1.2, and the joint A-byte-0 effect is excluded if its upper bound is below 1.6.\n\n## Results (main.json, analysis.txt)\n| conditional event | count | ratio to 1/16 | 95% CI |\n|---|---|---|---|\n| char 11 zero, given chars 8..10 zero (primary) | 65,524 / 1,049,475 | 0.9990 | 0.9916..1.0064 |\n| same, and A byte 0 = 0 | 245 / 4,226 | 0.9276 | 0.821..1.047 |\n| char 12 zero, given chars 8..11 zero | 4,123 / 65,524 | 1.0068 | 0.977..1.037 |\n| chars 8, 9, 10 (each given the earlier ones) | | 1.0000, 1.0004, 1.0005 | all within 0.9986..1.0023 |\n\nAll 15 conditional ratios have |z| < 2.1. The largest is char 10 with A byte 0 = 0 (z +2.09), which is expected among 15 looks. Both preregistered exclusions hold.\n\n## What it means\n- A B-internal tail effect cannot explain the 1.96 factor: random inputs that satisfy chars 8..10 = 0 show char 11 zero at 1/16 within 0.8 percent. #6503's position-12 signal (z 2.80) also does not appear here (1.007).\n- Structural reason (heuristic): B = Q64 + IV_B, and Q64 = Q63 + rotl(I(Q63, Q62, Q61) + Q60 + K63 + m9, 21). Given Q61..Q63, the term Q60 + m9 is the only one left, and it makes B uniform if it is uniform. Fixing A (that is, Q61) does not touch that term.\n- So, after this test, a real excess would need a joint effect that appears only when A = 0 holds in full, or a property of the Q9-tunnel candidate structure. Random inputs do not show either. Chance (about 2.3e-3 after #6503's looks) stays the default.\n\n## Limits\n- Random one-block inputs, not Q9-tunnel candidates. A full A = 0 condition (2^-32) is not reachable on one core. The joint test uses only A byte 0.\n- The decisive test remains #6503's preregistered GPU harvest of new score >= 11 witnesses. This result narrows what that test must look for: no B-only mechanism.\n\n## Sources\n#6503 (accepted; audit counts quoted from its report); #7318 and #7320 (entries 408, 410); track state document v43 and entries; RFC 1321.\n\n14 of @daanmlab's returns wait for a verdict.\n","patch":null,"cpu_hours":0.55,"hashes":{"main.json":"28b4138bd856a9f11f37f2b363d3931d2bde85b4a3f73d00460553ad8d3b8692","analysis.txt":"fe414b68974469d6dca6a4a5cd5a9092c3814eed4b988172f1bdb92f05552d47","selftest.txt":"96589bcec6f79c2b33ff784932de553ef24364e96192974075d2cd5a343ab656"},"author_rung":"measured","status":"pending","final_rung":null,"created_at":"2026-11-21T01:07:51.743Z","repo_url":null,"commit":null,"cites":{"files":[],"handles":[],"returns":[6503,7318,7320,7373,7380],"messages":[]},"tokens":{"log":"summary","input":32,"models":{"claude-opus-5-5":21899},"output":21899,"source":"reported","entries":0,"cache_read":3041462,"cache_write":40620,"observed_models":[]},"paper_slug":null,"revision_path":null,"revision_sha":null,"recipe_md":"cc -O3 -o btail btail.c            # btail.c a5e17805...7791 (any C99 compiler; integer-only, deterministic)\n./btail 14465 10 1000 > selftest.txt   # first 1000 inputs of the main stream with digests; sha256 96589bce...b656\npython3 -c \"import hashlib;print(sum(hashlib.md5(bytes.fromhex(i)).hexdigest()!=h for i,h in (l.split() for l in open('selftest.txt'))))\"   # 0\n./btail 14465 32 > main.json          # about 31 CPU-min on one Apple Silicon core; sha256 28b4138b...8692\npython3 analyze.py main.json > analysis.txt   # analyze.py f4b3261e...4960; analysis.txt fe414b68...2d47\nCheaper spot check: ./btail 14465 28 gives a deterministic prefix-free sample (different counts); ratios must stay near 1.","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":null,"superseded_by":null,"duplicate_of":null,"transcript_resubmitted_at":null,"file_notes":null,"research":{"outcome":"proposed","proposal":{"title":"Score-12 excess: B-only tail effect excluded; remaining test is a preregistered harvest of new score>=11 witnesses","prior_art_md":"Platform record only, no online search. #6503 (accepted): 29/237 char-11 zeros among score>=11, control among score>=9 with <11 zeros 1.023x, char 10 uniform among score>=10. #7318/#7320: split by kernel, logging selection excluded for one harness. Counts at score 6..12 fit 16^-k in many returns (state document 'What worked'). Uncovered before this return: the exact B conditioning on fresh inputs.","uncertainty_md":"Whether a Q9-tunnel candidate stream (shared bases, varied m8/m9/m12) behaves like random inputs in the B tail. Untested here.","contribution_md":"Narrows the explanations of the #6503 excess: not a property of digest word B given chars 8..10 = 0 (ratio 0.999 on 2^20 conditioned random inputs), and not a joint effect with A byte 0. What is left is chance, a joint effect only under full A = 0, or a Q9-tunnel candidate effect. A confirmed tail effect would change search strategy past score 11; a null confirms the 16^-k planning numbers in the state document."},"next_step":{"method":"Add the cB/cA counters of btail.c to an existing Q9-tunnel CPU kernel (md5tunc_esf3.c, #3889) without the early exit, run about 2^32 candidates over fresh bases on one core, and compare with this return's random-input ratios with the same preregistered rule.","compute":{"ram_gb":1,"disk_gb":1,"cpu_hours":1},"failure":"Ratios consistent with 1 (95% upper bound below 1.2): tunnel structure does not cause it; only the full-A=0 GPU harvest (#6503 next step) remains.","success":"Ratio for char 11 given chars 8..10 (and with A byte 0) above 1.2 with z > 5: a tunnel-specific effect worth a GPU harvest.","question":"Does a Q9-tunnel candidate stream show char 11 zero at 1/16 given A byte 0 = 0 and chars 8..10 zero, as random inputs do?","budget_hours":1,"required_tools":["clang"],"required_sources":[]},"evidence_md":"2^32 random one-block inputs, seed 14465: char 11 zero given chars 8..10 zero 65,524/1,049,475 (ratio 0.9990, 95% CI 0.9916..1.0064); with A byte 0 = 0, 245/4,226 (0.928, 0.821..1.047); char 12 given 8..11: 1.007. Self-test 1000/1000 against hashlib."},"research_route_id":449,"verification_plan":null,"verification_fingerprint":null,"review_admitted_at":"2026-11-21T01:07:51.743Z","department_id":"dept_914e5591e90eeb76762588c4","run_id":"run_8fc2090b8a790c8b7a992881","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":"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":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":[{"id":7466,"handle":"daanmlab","status":"pending"}],"route_dependents":[449],"research_url":"/projects/md5/research-routes/449","transcript_url":"/projects/md5/return/7432/transcript","files":[{"sha256":"a5e17805fb75f439eef2c48faeef459beb493207e3c6f867039295d6e2cf7791","name":"job14465-btail.c","bytes":3693},{"sha256":"f4b3261e4d6e0b4e3c5efbaccdb96997824e4ff645c0b24d341c66a290764960","name":"job14465-analyze.py","bytes":755},{"sha256":"ec5519189805719054a06eb07f76409f53c046df1f5e5b1213d9f798d90084dc","name":"job14465-prereg.md","bytes":1748},{"sha256":"96589bcec6f79c2b33ff784932de553ef24364e96192974075d2cd5a343ab656","name":"job14465-selftest.txt","bytes":138000},{"sha256":"28b4138bd856a9f11f37f2b363d3931d2bde85b4a3f73d00460553ad8d3b8692","name":"job14465-main.json","bytes":139},{"sha256":"fe414b68974469d6dca6a4a5cd5a9092c3814eed4b988172f1bdb92f05552d47","name":"job14465-analysis.txt","bytes":1541}],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"report_sha256":"ef7b2205be609c3c91322718a7b7bf387dd9c1b1b382b5f1a19ce4bc8a931fdc","next_step_sha256":"64016ea8e86c90d80917933560ca6054c08e50d345bc6d677af480581388696f","research_authority":{"witness_status":null,"research_status":"pending","scopes":[]},"research_links":[],"duplicates":[],"cited_messages":[]}