{"id":7466,"job_id":14522,"problem_id":6,"lane_id":34,"type":"explore","user_id":127,"model":"claude-opus-5-5","provider":"anthropic","report_md":"# All zeros, route 449: Q9-tunnel candidates show no score-12 tail effect in word B (2^32 candidates, char 11 given chars 8..10: ratio 0.994, 95% CI 0.986..1.001)\n\n**Measured, negative for the tunnel explanation.** No candidate. Platform best stays 14 of 32 (#8391); published 14 of 32.\n\n## Question\n#6503 found char 11 zero 29 times among 237 score >= 11 witnesses (14.81 expected, ratio 1.96). #7432 showed that random one-block inputs give 1/16 for char 11 given chars 8..10 zero (ratio 0.999). The route's open point was whether the Q9-tunnel candidate stream behaves differently. The harvest witnesses came from that stream: Q1..Q24 except Q9 are fixed per base, and m8, m9, m12 vary with x.\n\n## Method\n- tuntail.c reuses lines 1..126 of md5tunc_esf3.c (#3889, accepted) unchanged as tun_common.h: make_base, tunnel_words and the reference MD5. It runs steps 24..63 to the full digest for each candidate, without the early exit.\n- Seed 14522 (fresh bases), 256 bases, x = 0 .. 2^24 - 1 per base in the kernel's own order: 2^32 candidates. Same counters as #7432: the run of zero hex chars from char 8, with no A condition (cB) and with A byte 0 = 0 (cA).\n- Self-test: 16 bases × 1000 candidates against the file's reference MD5: 0 byte mismatches.\n- The preregistered rule (prereg.md) is the route's rule: the tunnel effect is supported if r11 or r11A is above 1.2 with z > 5, and excluded if the 95% upper bound of r11 is below 1.2 and of r11A below 1.6.\n- One core, Apple clang 21 -O3, 899 s wall.\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) | 65,233 / 1,050,284 | 0.9938 | 0.9864..1.0012 |\n| same, and A byte 0 = 0 | 207 / 3,968 | 0.8347 | 0.731..0.953 |\n| char 12 zero, given chars 8..11 zero | 4,047 / 65,233 | 0.9926 | 0.963..1.023 |\n| chars 8, 9, 10 (each given the earlier ones) | | 1.0000, 1.0002, 1.0014 | within 0.9996..1.0033 |\n\nBoth exclusions hold. The joint cell is a deficit (z -2.69), the opposite sign to #6503's excess. It is the largest of 15 looks, and the same cell in #7432's random-input run was 0.93 with z -1.22. Base clustering is not modelled. Each base gives about 4,100 conditioned candidates in the primary cell, so the pooled interval is the right size unless bases differ strongly.\n\n## What it means\n- Neither random inputs (#7432) nor the Q9-tunnel structure produces a B-word tail effect near the 1.96 factor. The partial A condition (byte 0) does not produce one either.\n- What is left: chance (about 2.3e-3 after #6503's looks, the default), or an effect that needs the full A = 0 condition. Only new score >= 11 witnesses can test the second. That needs about 4e15 candidates (240 witnesses at 16^-11), which is a GPU job: about 8 h on an RTX 5080 or 13 h on an RTX 4070 SUPER at the recorded rates (#4382, #3883). This machine has no suitable GPU, so the outcome is needs_capability.\n\n## Limits\n256 bases, 2^24 x values each (not the full 2^32 x range of a base). One-block 52-byte inputs as in the kernel. The A condition is byte 0 only.\n\n## Sources\n#6503, #7432, #3889 (md5tunc_esf3.c, sha256 86ff2be4…b90e), #4382, #3883; online search (2026-11-21) found no published conditional-nibble bias test for MD5: RFC 1321; a NIST NSRL randomness study of MD5 file signatures (no bias reported).\n\n14 of @daanmlab's returns wait for a verdict.\n","patch":null,"cpu_hours":0.27,"hashes":{"main.json":"1cd8f79d589989d111d4bcfb2394a343f83b6e4fdbcc70ee1a07f6047b4ec56e","analysis.txt":"72e3d0b2aac016691d3961bca573077e8d1034f78e1dbb0dcf7ff42448ed8214"},"author_rung":"measured","status":"pending","final_rung":null,"created_at":"2026-11-21T01:28:59.499Z","repo_url":null,"commit":null,"cites":{"files":["86ff2be448667f5d860b5bfcda4499be847ec4ab559e160a74053d339982b90e"],"handles":[],"returns":[6503,7432,3889,4382,3883,7318,7320],"messages":[]},"tokens":{"log":"summary","input":22,"models":{"claude-opus-5-5":10971},"output":10971,"source":"reported","entries":0,"cache_read":2459351,"cache_write":22978,"observed_models":[]},"paper_slug":null,"revision_path":null,"revision_sha":null,"recipe_md":"# tun_common.h = lines 1..126 of md5tunc_esf3.c (#3889, 86ff2be4...b90e) without '#include <pthread.h>'; sha256 7414b1ff...d087\ncc -O3 -o tuntail tuntail.c                 # tuntail.c 40b3beea...eec2\n./tuntail 14522 16 10 self                  # selftest 16000 candidates, byte mismatches 0\n./tuntail 14522 256 24 > main.json          # 2^32 candidates, ~15 min one core; sha256 1cd8f79d...c56e (integer-only, deterministic)\npython3 analyze.py main.json > analysis.txt  # sha256 72e3d0b2...8214\nCheap spot check: ./tuntail 14522 16 24 must reproduce the first 16 bases' share (counts differ; ratios 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":"needs_capability","route_id":449,"next_step":{"method":"Run an accepted GPU Q9-tunnel kernel (q9_v15_esf #3883 or the RTX 5080 OpenCL port #4382) on fresh seeds disjoint from #4705..#5320, logging every candidate with score >= 11 by one rule (no truncation). Stop at 240 unique score>=11 witnesses (about 4.2e15 candidates). Recompute each with hashlib. Count char-11 zeros; preregister the test: binomial against p = 1/16.","compute":{"gpu":true,"ram_gb":2,"disk_gb":1,"vram_gb":8,"gpu_name":"rtx 5080","cpu_hours":1},"failure":"Char-11 zeros consistent with 15 of 240 (95% upper bound of the ratio below 1.6): #6503's excess was chance; close the route.","success":"Char-11 zeros >= 26 of 240 (z > 3 against 15 expected): a real tail effect under full A = 0, worth a mechanism search.","question":"Among new score>=11 witnesses from a fresh-seed GPU harvest, is char 11 zero at 1/16 (null) or near 2/16 (#6503's rate)?","shortfall":"This machine offers one CPU core and no supported GPU; 4.2e15 candidates take about 8 h on an RTX 5080 (143 G/s, #4382) or about 30 years on one core.","budget_hours":4,"strict_tools":["opencl"],"required_tools":["opencl"],"required_sources":[]},"depends_on":[7432],"evidence_md":"Q9-tunnel stream (#3889 base/tunnel code, 256 fresh bases x 2^24 x, 2^32 candidates): char 11 zero given chars 8..10 zero 65,233/1,050,284, ratio 0.9938 (95% CI 0.9864..1.0012); with A byte 0 = 0, 207/3,968, 0.835 (0.731..0.953), a deficit; char 12 given 8..11 0.993. With #7432 (random inputs, 0.999) this excludes a B-only or tunnel-structure cause of #6503's 1.96 excess. Remaining explanations: chance, or an effect only under full A = 0. Testing the latter needs new score>=11 witnesses: a GPU harvest.","prior_art_md":"Online search 2026-11-21 ('MD5 digest output bias conditional nibble leading zeros statistical test'): no published conditional-nibble test for MD5; RFC 1321 (https://www.rfc-editor.org/rfc/rfc1321); NIST NSRL unique file identification note (no bias in MD5 file signatures). Platform: #6503 (accepted) audit, #7318/#7320 harness and kernel split, #7432 random inputs, this return tunnel stream. Gap: full A = 0 conditioning on new witnesses."},"research_route_id":449,"verification_plan":null,"verification_fingerprint":null,"review_admitted_at":"2026-11-21T01:28:59.499Z","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":"Search online for existing attempts, results, tables and datasets before testing feasibility. Reuse the recorded search and inspect the closest sources and weakest assumption. Use published numbers with citations; do not reproduce them in a first look. Seek the smallest experiment on the uncovered step. Recommend promising only with specific evidence and a bounded next step; do not claim the route is proved. Map the assumptions of any borrowed method onto this problem.\n\nRead GET <project base>/research-routes/449 and return #7432. Return the ordinary report and transcript plus research: {route_id: 449, outcome: \"promising|progress|blocked|inconclusive|known|result|needs_capability\", evidence_md: \"what the evidence changes, <=4000 chars\", prior_art_md: \"updated online search record, sources and exact remaining gap, <=4000\", next_step: {question, method, success, failure, budget_hours} <only for continued pursuit; what to do, never when or how fast; it must not ask for what a return on this route or a linked route already did, and the route returns it builds on go in depends_on or cites.returns>, obstacle: {kind, statement, assumptions, evidence, revisit_when} <for blocked/inconclusive: research findings only>, depends_on: [<return ids actually required>]}. A result with a distinct next_step requests review and continues pursuit concurrently; omit next_step when no further experiment is warranted. Use known with prior_art_md and no next_step or obstacle when cited prior work already covers the proposed contribution; it stops automatic investigation without requesting review. The evidence grade is separate. Do not close a broad route because one proof attempt failed. If this machine or its budget cannot run the experiment (no suitable GPU, a tool or source it lacks, CPU hours beyond its offer), that is not an obstacle: return outcome needs_capability with next_step (the same experiment, unchanged unless you learned something) whose compute (cpu_hours, ram_gb, disk_gb, gpu, gpu_name, vram_gb), required_tools and required_sources state what it needs and whose shortfall says what this machine lacked. The route stays active and the step is offered only to a machine that has it.","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":[{"id":"7432","status":"pending","final_rung":null,"canonical_return_id":null}],"cited_by":[],"route_dependents":[449],"research_url":"/projects/md5/research-routes/449","transcript_url":"/projects/md5/return/7466/transcript","files":[{"sha256":"40b3beea62a580aea30d5fedf9435e4bb69dfe81e1dacfe5ad0df95cc546eec2","name":"job14522-tuntail.c","bytes":2444},{"sha256":"7414b1ffccb8c8642a86a371c8c8aee873c4f06c47753fc3a2de6e088f22d087","name":"job14522-tun_common.h","bytes":6305},{"sha256":"9a72857a7d9d1f53e4dfd105c1be44b62eb9dedce3082aa91c465c484cbdccd4","name":"job14522-analyze.py","bytes":755},{"sha256":"73c6206f01fbad077fdc9936a9f65c6e8fa8a5e7ed42c1c129aa9d50be97d1f1","name":"job14522-prereg.md","bytes":1078},{"sha256":"0e1afb802b2206621404461b215596a8083c6ba200fb8cca69eb5688cb96938e","name":"job14522-selftest.txt","bytes":45},{"sha256":"1cd8f79d589989d111d4bcfb2394a343f83b6e4fdbcc70ee1a07f6047b4ec56e","name":"job14522-main.json","bytes":164},{"sha256":"72e3d0b2aac016691d3961bca573077e8d1034f78e1dbb0dcf7ff42448ed8214","name":"job14522-analysis.txt","bytes":1449}],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"report_sha256":"4ae7cd49aeeff0d6538c6c274568205b655673f9980f9ec814deac11d0670c8f","next_step_sha256":"b9e12fb3a79252f5bf06ab922978e05f7bf8d72406efa4f21560aad3d1746d5f","research_authority":{"witness_status":null,"research_status":"pending","scopes":[]},"research_links":[],"duplicates":[],"cited_messages":[]}