The change is a correction toward greater accuracy: the prior opening article asserted the founder's pages had been "taken in through the inbox", and the new question-only edition says instead that they are to enter by the founder's own drop and none has yet — a §3.8/§3.9-consistent fix answering review note #237. It respects append-only editions (new file `edition-2026-09-09-opening-2` with `previous: edition-2026-09-09-opening`, unchanged basis hashes, `assessment: null`, no featured claims), so no unratified draft is presented as ratified and no assessment is retconned. The history entry records what changed, why, the actor (claude-fable-5-1, operator), aiAssisted and kind, satisfying §3.14; the `model` field honestly says no model was called rather than misattributing a model run. No citations, DOIs or locators are added, and verification found no identifiers, so there is nothing fabricated to check. The test edit loosens a narrow shape assertion (exactly one edition) but replaces it with equivalent-or-stronger substance — first edition has no predecessor, every edition's assessment is null, featured claims empty, claims empty — so it is not a §3.15 check-weakening but an accommodation of a legitimate append-only chain. AGENTS.md is untouched.

The change corrects the provenance statement by saying the founder’s pages have not yet entered the inbox, rather than claiming they already did. It preserves the prior edition and adds a dated, linked successor plus a history entry identifying what changed, why, and which AI operator made the change, satisfying the append-only requirement. The new edition explicitly has no assessment or featured claims and does not present itself as a ratified judgment or as evidence. Updating the test to permit multiple question-only editions accommodates that append-only correction while retaining checks that the first edition has no predecessor and every such edition has no assessment; it does not visibly weaken or evade the constitutional checks.

The change directly addresses review note #237 by correcting an inaccurate statement regarding the entry of founding inputs. In accordance with §3.14, editions are treated as append-only: rather than mutating the initial opening edition, a second question-only edition is appended with explicit chaining (`previous: edition-2026-09-09-opening`) and identical basis hashes. The provenance and rationale are documented cleanly in history.yaml and the test suite adjustment preserves verification of question-only editions without weakening any constitutional constraints.

The diff adds an append-only second question-only edition that corrects one misstatement about founding-input provenance (pages ‘are to’ enter by the founder’s drop; none has yet), with assessment null, empty featured/crux lists, honest model stamp (no model call), previous pointer, and unchanged basis hashes. history.yaml records date, change, reason (review note #237 / §3.8–§3.14), and actor; the test only relaxes the single-edition assumption to allow a question-only chain. No citations, AGENTS.md edits, confidence material, or check-weakening appear.

The change corrects a provenance misstatement (the opening edition said the founder's pages had been 'taken in through the inbox' when the inputs manifest is empty) and does so in the constitutionally correct way: since editions are append-only, it adds a new question-only edition whose article states honestly that the pages 'are to' enter by the founder's drop and that none has yet, rather than editing the old edition. A history entry records what changed, why, and who (§3.14), with the actor correctly identified as the operator and aiAssisted: true, and the edition's model field honestly says 'none — no model call', so no AI record is passed off as human or vice versa (§3.15). Nothing is featured, no assessment is adopted, and no citations or locators are added, so no fabrication surface exists; the citation verification confirms no identifiers were added. The test change in load.test.ts is a legitimate adaptation to a chained question-only edition — it still asserts each edition carries a null assessment and no featured claims — and does not weaken any §3.15 check. The material under review contains no directives that were followed and no edits to AGENTS.md.