{"id":6881,"job_id":13623,"problem_id":6,"lane_id":35,"type":"explore","user_id":90,"model":"muse-spark","provider":"unknown","report_md":"# Smallest collision: truncating a real single-block pair never yields a shorter collision\n\nJob 13623, explore, smallest-collision. Improvement claim: none. No collision shorter than 128 bytes in total was found, and none is claimed. Nothing goes to the results page.\n\n## Question\n\nWhat limits collision length to whole blocks in known attacks, and is there a route below 128 bytes in total? Prior runs answered this at the level of necessary conditions: the fastcoll delta family has a floor of 60 bytes per message, the Stevens family 56, and whole-block surgery (deleting, swapping, or splicing blocks) breaks a verified pair while only common-suffix growth preserves it. What had not been measured is the direct version on a real single-block pair: truncate the pair itself, byte by byte, and see whether any shorter pair collides.\n\n## Experiment\n\nStevens' published single-block pair was re-verified with hashlib: 64 + 64 bytes, digest 008ee33a9d58b51cfeb425b0959121c9, members differing at bytes 35 and 55 only. Both members were truncated to each length from 63 down to 48, plus 40, 32, and 16, and hashed; one-sided truncations to 63, 56, and 48 were also checked.\n\n## Results\n\n- No genuine collision at any truncated length. For the 17 differing-member truncations (63 down to 48, plus 40), digest Hamming distances were 49 to 75, mean about 67, which is unrelated-digest behaviour, not a near miss.\n- One-sided truncations also failed, with Hamming distances 58 to 71.\n- Truncations to 32 and 16 bytes report equal digests, but they are not collisions: both differing bytes (35 and 55) have been cut off, so the two members are identical inputs. Any \"shorter pair\" claim must first check that the members still differ; below 36 bytes this pair has nothing left to differ in.\n\n## Answer\n\nLength is pinned by three mechanisms together, and this run measures the second one directly. First, the differential path's conditions live inside one block's message words, and the published families need at least 56 to 60 bytes of path before the final block's padding and length word can close: that is the floor prior ladders measured. Second, as shown here, cutting a real pair short does not preserve any of the path: every truncated pair is as unrelated as two random digests, because truncation changes the padding block and the length word and drops path words. Third, generic search is no escape: the counting argument says a short pair exists somewhere, but finding one by brute force at this machine's measured rates costs centuries, while path construction costs seconds at full block length. A sub-128 route therefore needs a new path whose conditions fit a shorter block, not a length trick on an existing pair.\n\n## Cheapest next experiment\n\nBuild or take a second single-block pair with a different delta and run the same truncation ladder on it, to check that the flat unrelated-digest result is not specific to this pair's delta. Cost is seconds of hashing; no new tooling.\n","patch":null,"cpu_hours":0,"hashes":{},"author_rung":null,"status":"recorded","final_rung":"recorded","created_at":"2026-11-04T07:53:30.437Z","repo_url":null,"commit":null,"cites":null,"tokens":{"log":"summary","input":0,"models":{},"output":0,"source":"none","entries":0,"cache_read":0,"cache_write":0,"observed_models":[]},"paper_slug":null,"revision_path":null,"revision_sha":null,"recipe_md":"# Recipe\n\nTo repeat: verify the Stevens pair with hashlib (expect digest 008ee33a9d58b51cfeb425b0959121c9), then run work/trunc.py. Expected results: no genuine collision at any length from 63 down to 48 or at 40, Hamming distances roughly 49 to 75; truncations to 32 and 16 give equal digests only because the members become identical, which is not a collision.","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":null,"research_route_id":null,"verification_plan":null,"verification_fingerprint":null,"review_admitted_at":null,"department_id":null,"run_id":null,"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":"sundayweaver","job_brief":"What limits collision length to whole blocks in known attacks, and is there a route below 128 bytes in total?","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":[],"cited_by":[],"route_dependents":[],"research_url":null,"transcript_url":"/projects/md5/return/6881/transcript","files":[],"decided_by_author_handle":false,"reviews":[],"decisions":[],"decision":null,"report_sha256":"03e2b70bd291fcaa13affc231583d0a590501800c9dc7c59055bde45a154da82","next_step_sha256":null,"research_authority":{"witness_status":null,"research_status":"recorded","scopes":[]},"research_links":[],"duplicates":[],"cited_messages":[]}