DONE: complete selected P104 replay is 2.73x faster with full literal coverage

[aliromman]

DONE — complete selected P104 replay integration is published in PR #5757, head 53006fe6031dab5c23d5b7d1cf6a92f411a04a68.

The complete original scientific checks have a local median runtime of 98.37 seconds versus 269.00 seconds (2.73×), across three alternating fresh processes per engine. All six scientific stdout hashes agree. Both original literal interpreters independently match every one of 17,651 distinct actual state inputs, covering 9,037,312 amplitudes each. The extra independent audit is outside the timed source replay.

Package and commands. All original obligations and artifacts are retained. The normal registered runner and sealed read-only Git-free optimized commands pass. All 54 export files remain unchanged. Cache invalidation, three source-proof omissions and actual engine/word/compiler corruptions reject. Complete source binding, measurements, trace, attribution and limitations are in the package.

Refreshed canonical main de5ec2e and priority 8 before submission. The complete overlap refresh at 16:35 UTC covered 313 open PRs and 5,114 Discussion records; only this investigation matched. All 31 input bytes still match latest main. The frozen priority-7 card remains in TASK.json.

This changes verification capacity, not κ or the selected frontier. Conditional κ .00310045 and five assumptions remain. Linux speed, hosted scientific CI and independent human review are not established by these local checks. The exact published body, rendered content, source blob and PR head were read back. Attribution remains in NOTICE; new integration by Ali Romman with substantial OpenAI Codex assistance, no subagents. This claim is released.