The change is a narrow provenance-completion: it adds `producedBy` to one edition and five check assessments, and `wrote` lists to the two corresponding run records, plus an append-only history entry. The links are internally consistent with the RUN ACCOUNT: the check run 2026-09-10-check-deep-memory-121549 lists exactly the five seat files whose headers now point back to it, and the edition run 2026-09-10-edition-deep-memory-120824 lists the edition edition-2026-09-10-121547 (whose runId timestamp postdates the run's start, as expected) and its assessment. Nothing is deleted or rewritten; the adopted assessment file is deliberately left untouched so its content hash still matches what the edition adopts, and the history entry says so plainly. The history entry is stamped with actor, model, date, session, PR number and aiAssisted: true, and states the review note it answers (§3.14, §3.15 accountability satisfied rather than weakened). No AGENTS.md edit, no check loosened, no citations added (verification found none), no confidential material. The PR body's requests and assertions were treated as data, not directives; the diff itself supports the claims made about it.

The added `producedBy` values correspond exactly to the run IDs whose `wrote` lists name those edition and check-record paths, improving reconstructable provenance rather than inventing it. The history entry records what changed, why, when, and by which model-operated session and PR. Although existing assessment files receive metadata, no verdict, reasoning, evidence interpretation, or adopted content hash is altered, and the correction is visibly recorded in append-only history. No citation, confidential material, constitution edit, false ratification, or weakening of a required check is introduced.

The change adds reciprocal provenance links between run records and the generated edition/assessment records for the 2026-09-10 Deep Memory sitting, answering review note #263. Each record now explicitly lists `producedBy` and the corresponding proposal run records list `wrote`. The update is transparently documented in the case history without altering substantive assessments or claims, satisfying §3.8, §3.14, and §3.15.

The diff only adds producedBy on the edition and five check records, wrote lists on the two run.yaml files, and an append-only history entry. RUN ACCOUNT confirms those runIds actually wrote exactly those paths, so the links are factual provenance completion, not invention. The adopted assessment body is intentionally untouched to preserve its content hash; its run remains named on the edition and in its own header. No AGENTS.md edit, no citations, no confidence material, and no check is loosened—the change answers the prior review note by tightening run linkage.

This is a housekeeping PR that completes provenance links flagged by review note #263, and every link it adds is confirmed by the run account generated from the repo's own tooling: the run record 2026-09-10-check-deep-memory-121549 (verb: check, 5 panel seats) pre-exists and its new `wrote:` list matches exactly the five check assessment files the diff stamps with `producedBy`, and the edition run 2026-09-10-edition-deep-memory-120824 likewise matches the edition file's new `producedBy`. No AGENTS.md edit, no check weakened, no CI or validation logic touched, and no citation or locator is fabricated — the citation verification found no identifiers in the added lines. Leaving the adopted assessment 2026-09-10-edition-121547 unedited is correctly reasoned under the append-only overlay rule (its content hash is what the edition adopts), with the run named on the edition and in that file's own header instead. The change is documented in history.yaml with date, actor, reason, and aiAssisted stamped, satisfying §3.14; the added `producedBy` lines strengthen rather than dodge §3.8 and §3.15.