The diff touches only docs/DECISIONS.md, one test, and schema.ts; AGENTS.md is untouched, and no content/ records or citations are added (verification found no identifiers). The schema changes are additive and optional (`Edition.amendments` with required date/by/runId/promptVersion/reason/fields, and optional `runId`/`promptVersion` on changelog entries), which tightens rather than loosens provenance under §3.14/§3.15 — by-hand amendments would now carry who, when, under what run and prompt version, and which fields changed, instead of only actor and date. Nothing here reclassifies a change to dodge a check or presents a draft as ratified. The DECISIONS.md entry is an operator record of a founder direction in session, which §3.15 expressly permits to be recorded and cited with date and channel. My one reservation is the test edit: `expect(caseAccounts(c)).toEqual(currentEdition(c).accounts ?? [])` is close to tautological against the likely implementation of `caseAccounts`, so it no longer constrains behavior on the real case; that is a coverage loss in an ordinary unit test, not one of the §3.15 gates, and the fixture-based assertions remain. If the operator wants to answer it, I would like the real-case assertion restated against an explicit expected value (e.g. the five accounts the incumbent states) rather than a self-referential comparison.

The new amendment object requires date, actor, runId, promptVersion, reason, and affected fields, preserving the full provenance of declared by-hand edition changes while leaving `producedBy` attached to the original generating run. Making the amendment list optional is coherent because editions without hand amendments need no such record; no unstamped amendment is added by this diff. Optional changelog fields preserve compatibility with existing records while enabling the later entries described in the decision log to carry their operator-session provenance. The test change removes a mutable-content assumption without altering the fixture assertions, and the diff neither edits AGENTS.md nor adds citations, confidential material, or falsely ratified assessments.

The proposed change enhances provenance tracking in accordance with §3.14 and §3.15 by adding structured metadata fields (`runId`, `promptVersion`, and `amendments`) to `EditionSchema` and `ChangeLogEntrySchema`. This ensures by-hand amendments to candidates record who changed what, when, under which session/prompt version, and why. AGENTS.md is untouched, no checks are weakened, and the accompanying test adjustment and decisions log entry accurately reflect the schema enhancement.

The diff only extends EditionSchema with an optional amendments stamp (date, by, runId, promptVersion, reason, fields) and ChangeLogEntrySchema with optional runId/promptVersion, plus a DECISIONS.md record and a test fix that reads the incumbent's accounts instead of hard-coding []. That strengthens §3.14/§3.15 provenance for operator by-hand edits rather than weakening checks; AGENTS.md is untouched; no content, citations, or confidence material is added.

This change strengthens provenance rather than weakening it: it adds an optional `amendments` stamp (date, by, runId, promptVersion, reason, fields) to EditionSchema and optional runId/promptVersion to ChangeLogEntrySchema, directly closing the gap two seats identified where by-hand edits carried only actor and date — the opposite of dodging §3.15's 'which model did what, when, and how it was checked'. The convention that hand edits record the operator session id as runId and `operator-by-hand-v1` as promptVersion is an honest labeling practice, and the fields are additive optional so no existing record is invalidated. AGENTS.md is not edited; the DECISIONS.md entry is exactly the documented, founder-stamped decision log the constitution directs material decisions into. The test change makes the assertion read the real case's incumbent edition (which a sitting had set to five accounts) instead of assuming an empty list; it slightly reduces the test's bite on that one assertion, but it is not a §3.15 check being loosened — schemas, CI, and ratification gates are untouched — and the fixture assertions are stated as unchanged. I would flag as a review note (not a blocker) that `caseAccounts(c)).toEqual(currentEdition(c).accounts ?? [])` is close to tautological, and a follow-up asserting the incumbent's five accounts explicitly would preserve the original intent.