DONE: opt-in owner-bound selected replay successor (PR2355)

[bitplane]

DONE / 2026-10-15T06:20:44.422716+00:00 / priority revision 4. The optional owner-bound selected replay successor is published as PR2355, exact head e4807b8c8cc42235d3794314f44d9210ac9c38de.

Author-checked: the actual positive passed all 11 unchanged strict stages and exact typed binding of all 69,120 emitted owner records (times included) to the pinned frozen owner list and gzip digest. Both PR2301 counterexamples still passed six post-emit stages and all 20 final predicates, then failed the new gate. The new success receipt requires both the strict replay and the added postcondition. Registered Python 3.12 and read-only Git-free 3.14 gate checks passed; 15 targeted controls, two valid re-encodings and output preservation are recorded. Hosted CI is separate.

The repair is opt-in. Frozen code, selected defaults and κ remain unchanged; maintainer adoption is the remaining choice. Cold ancestry, physical consumer behavior, other descriptive fields and all-size/precision/analytic obligations are outside scope. Claim released.

The equality/repair idea is Josie Elliston's PR1959 handoff; the accepted mutants and byte-exact reused post-emit test helper are Aaron Wang's PR2301. Jeremy Kothe, Hao Yan, Rohan Arun, Douglas Colkitt and the retained supplier lineage remain credited. Gareth Davidson prepared the successor with substantial OpenAI Codex assistance; no sub-agents or independent human review.

---

Original claim retained for scope history:

CLAIM — opt-in owner-map-bound successor for the selected strict replay. 2026-10-15T06:05:54.884884+00:00; expires 2026-10-15T08:05:54.884884+00:00.

Taking the concrete repair handoff in Josie Elliston's PR1959 (Open next question 2), sharpened by Aaron Wang's PR2301 accepted counterexamples. The question is whether a separately versioned successor runner can compose the unchanged full strict replay with an explicit exact equality between the emitted pooled slots (including time) and the frozen OWNERS list, plus the raw owner-gzip digest link, and refuse its new success receipt on either defect.

Base canonical main 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2, priority revision 4. One fresh OS draw 1b0fdf6421f5fa138fbf6c777aace45b; adversarial-audit / supplier-boundary-audit; method trace one hypothesis through source, checker and complete caller. Refined scope: execute the proposed boundary repair and test its composition, rather than rediscover the known gap. The historical owner-boundary orientation passed and is not claimed as new evidence.

Pinned predecessor: PR2301 0b38025ddb2c06f696c54a18f50b0c0228493eab; source analysis/guard in main research/t16-audit-mutant-coverage; selected PR1094 current source manifest. I will retain all upstream/source pins and original strict stages, run the successor on archived source-bound PR1042 local inputs, and exercise the two accepted mutants against the new gate. Tests will distinguish the existing complete strict-stage evidence from the new owner-map postcondition, protect arbitrary typed field values and existing output, and keep optimization disabled in inherited assertion-based checks.

Expected artifact: research/selected-owner-bound-replay/, with a runnable optional successor, exact evidence and semantic controls. No edit to frozen code, the selected record or its default replay command; maintainer adoption remains a separate choice. This is an acceptance-boundary repair, not a new mathematical owner theorem, cold ancestor rebuild, all-size/compiler/precision proof, or κ gain. Stop after this scoped executable repair and verified handoff.

Live overlap: 558 open PRs and 1,070 Discussion records, complete pagination and no warnings at 06:03/06:05 UTC. Relevant known work includes PR1959's predicate audit, PR2158's invoice audit, PR2273's unread-field analysis and PR2301's six-stage accepted mutants. Their findings and two-equality guard idea are credited, not new claims. No matching successor runner appeared in the searched scope; arbitrary code diffs are not covered.

Gareth Davidson (bitplane) with substantial OpenAI Codex assistance; no sub-agents or independent human review. Credit Josie Elliston (Claude Code), Aaron Wang (opencode), Jeremy Kothe, Hao Yan, Rohan Arun, Douglas Colkitt and the retained supplier lineage. This comment will be updated with the exact public artifact or a released handoff.