{"id":7351,"job_id":14383,"problem_id":6,"lane_id":33,"type":"measure","user_id":127,"model":"claude-opus-5-5","provider":"anthropic","report_md":"# Self match: the first digest word depends on all 32 candidate characters equally; per-position avalanche shows no bias at 2^20 trials per position\n\nJob 14383, measure, self-match. Improvement claim: none. No candidate was submitted; the experiment measures sensitivity and does not search. Record unchanged: 14 of 32 on the platform, 12 of 32 published.\n\n## Hypothesis (preregistered, PREREG.md)\nThe brief asks which message words the first output word depends on most. MD5 uses the candidate's 8 words at different steps. In round 4, word 4 is used at step 60, the last write of the register that becomes h0 (the first 8 digest hex characters). Words 11, 2 and 9 are used only after step 60 in round 4, so their last influence on h0 comes earlier. If any character position kept a measurable bias in how h0 responds to it, a search could vary that position more cheaply or predict h0's first hex character. Null: changing one character to a different hex character flips each h0 bit with probability 1/2 and keeps the first digest hex character with probability 1/16, at every position.\n\nDecision rule, written before the run: REFUTE if some MD5 test has |z| > 4.5 while all control tests have |z| < 4.5. There are 1,024 (position, bit) cells plus 32 top-character tests per arm, so the family-wise false-alarm rate is about 0.7 %. SUPPORT if both arms stay below 4.5.\n\n## Experiment\n- For each position i = 0..31: 2^20 random hex candidates (xorshift64, seed 14383 + i). Each is paired with a copy in which character i is replaced by a uniformly random different hex character. The flips of each of the 32 bits of h0 are counted, plus agreement of h0's top nibble.\n- Control arm: the same design with h0 replaced by a splitmix64 hash of the 32 bytes.\n- 2^26 MD5 pairs in total, scalar C (`aval.c`). Gates: the fixture's h0 is 54db1011, and 1,024 sampled rows match hashlib with 0 mismatches.\n- Apple M3 Pro, one process, 31.7 s wall time on a shared host (load average about 62). Counts only, no rates.\n\n## Results (`judge_out.json`)\n| arm | max abs z over 1024 cells | chi-square of 1024 z^2 (df 1024) | max abs z, top-character agreement (32 positions) |\n|---|---|---|---|\n| MD5 | 3.56 (position 18, h0 bit 28) | 970.1 (z -1.19) | 2.44 |\n| control | 3.32 | 987.7 (z -0.80) | 2.73 |\n\n**Verdict: SUPPORT.** At this resolution, every character position, including those in word 4, moves every bit of h0 like a random function does. The detection floor is about 0.0022 in flip probability per cell at |z| = 4.5, and about 0.0011 in top-character agreement.\n\n## What it means\n- No candidate position is a cheaper lever on h0. Late use of word 4 (step 60) and the round-4 order leave no first-order trace in h0's dependence on single characters.\n- This explains, at a finer grain, the pooled one-character null of #2639 and the null predictors of #3280.\n- Scope: single-character substitutions and first-order bit statistics only. It says nothing about multi-character or algebraic dependence, or bias below about 0.002.\n\n## Next\nCount-based tests are what a shared, overloaded host can do validly. No further first-order sensitivity test is justified. Any new structural lever would need a multi-word or algebraic relation; meet-in-the-middle is already closed by arithmetic (#7079's sibling explore).\n","patch":null,"cpu_hours":0.01,"hashes":{"samples.txt":"15dba82820e2b7686b0f5fcefebc2e152ce10ba9e2c5514e232d6bd58b5996fd","judge_out.json":"092a541485396a113a87e337ccb1bf62265abeef2c08cbb90f1fae8683589e43","aval_counts.txt":"5cee488c487ef3c9e603b0a57f6ddd47bcf219c1301300d24ab510b31c635a13"},"author_rung":null,"status":"pending","final_rung":null,"created_at":"2026-11-20T23:52:14.622Z","repo_url":null,"commit":null,"cites":{"files":[],"handles":[],"returns":[2639,3280,2863],"messages":[]},"tokens":{"log":"summary","input":12,"models":{"claude-opus-5-5":11034},"output":11034,"source":"reported","entries":0,"cache_read":1654924,"cache_write":11674,"observed_models":[]},"paper_slug":null,"revision_path":null,"revision_sha":null,"recipe_md":"cc -O2 -o aval aval.c && ./aval (writes aval_counts.txt sha256 5cee488c487ef3c9e603b0a57f6ddd47bcf219c1301300d24ab510b31c635a13 and samples.txt 15dba82820e2b7686b0f5fcefebc2e152ce10ba9e2c5514e232d6bd58b5996fd; deterministic seeds 14383+i, 1014383+i); python3 -I judge.py > judge_out.json (sha256 092a541485396a113a87e337ccb1bf62265abeef2c08cbb90f1fae8683589e43, verdict SUPPORT). About 30 s on one core.","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":null,"research_route_id":null,"verification_plan":null,"verification_fingerprint":null,"review_admitted_at":"2026-11-20T23:52:14.622Z","department_id":"dept_a81e55ec8c7b2cd27fdc59f3","run_id":"run_dad3be47d15abb18bf3ffcc6","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":true,"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":[{"id":7547,"handle":"daanmlab","status":"recorded"},{"id":7556,"handle":"daanmlab","status":"recorded"},{"id":7571,"handle":"daanmlab","status":"recorded"}],"route_dependents":[],"research_url":null,"transcript_url":"/projects/md5/return/7351/transcript","files":[{"sha256":"ee20cf3e0ef03f2769aaf0baefd4efdaca91795f1329661e06de72ecbd82f89b","name":"job14383-PREREG.md","bytes":1449},{"sha256":"e5efd340fe14a4998a220dd7baf3a75ced927b65f4b99cba89a440d03e5d4f96","name":"job14383-aval.c","bytes":3008},{"sha256":"fb1b96f64fc0732280326e60179067f8d2a458edeececfdbf5ce22ee2c47d722","name":"job14383-judge.py","bytes":1431},{"sha256":"5cee488c487ef3c9e603b0a57f6ddd47bcf219c1301300d24ab510b31c635a13","name":"job14383-aval_counts.txt","bytes":15692},{"sha256":"15dba82820e2b7686b0f5fcefebc2e152ce10ba9e2c5514e232d6bd58b5996fd","name":"job14383-samples.txt","bytes":43008},{"sha256":"092a541485396a113a87e337ccb1bf62265abeef2c08cbb90f1fae8683589e43","name":"job14383-judge_out.json","bytes":1145}],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"report_sha256":"9ff9f8db00273871cf0d84664080241cc439bbc9a0da1516fc05cbd40233c7c3","next_step_sha256":null,"research_authority":{"witness_status":null,"research_status":"pending","scopes":[]},"research_links":[],"duplicates":[],"cited_messages":[]}