[kim-em]

The complete primary SignDet Phase-6 comparison now has a result: 29 workloads pass, while direct joint checking at size 127 and reduced joint production at size 255 report regressions. Retain the complete 31-workload, four-block capture and the consumed resolution reservation, and replace the reports’ stale “comparison incomplete” wording with these two unresolved findings. All original observations and earlier failed/interacting captures remain intact; no phase counter changes.

Validation: 850 compressed/original-byte bindings checked; all 248 exports and 1,256 timed rows checked for matching answers; final source/tool/checker integrity checked. The hash-bound native checker reproduces all 31 saved reports and exit codes without new timings. All 107 relative documentation links and whitespace checks pass. No new measurements, source changes, or unchanged retry.

Refs #10575. Independent Opus review starts immediately after opening; CI will run on this exact head.

[kim-em]

The required fresh isolated Claude Opus process was launched immediately after opening this PR. It exited1 with: “You've hit your weekly limit · resets 4pm (UTC)”. No review was obtained and no review acceptance is claimed. The PR is now draft pending that mandatory review and successful exact-head CI. The complete capture/bindings, all248exports/1256rows and all31 native-checker replay checks pass locally; no new timings or automatic retry was performed.