The change is narrow engineering: `ledgerHash` now excludes the research agenda's lifecycle fields (`status`, `statusNote`, `statusBy`, `statusDate`), mirroring the existing exclusion of source locator fields, plus a test asserting a settled item does not move the hash, plus a DECISIONS.md entry recording the rationale. No content/ records, citations, locators, or assessments are touched, and the citation tooling found no identifiers to check. AGENTS.md is untouched, so §3.15's reservation is not implicated. The hash governs only when the scheduler owes a case a new edition; it is not one of §3.15's gates (fail-closed CI schema/dangling-ID/citation checks, multi-model ratification, parking of contested changes), so excluding the edition verb's own bookkeeping is not check-weakening — the status change remains visible in the case history and agenda page, satisfying §3.14's versioning requirement. The DECISIONS.md entry is stamped with date and channel and attributes the direction to the founder in session, which the constitution expressly permits as operating record. Minor nit only: the test's `as typeof slice` cast slightly strains §8's avoid-loose-typing spirit, which is not a violation.

The change narrowly excludes four post-edition research-lifecycle fields from the scheduler’s ledger hash while leaving claims, evidence, sources, substantive research fields, studies, and images hashed. Those fields remain versioned in git; nothing in the diff deletes their provenance, validation, or public records, so §3.14 remains satisfied. This does not weaken the fail-closed checks or concurrence gates named in §3.15; it prevents an edition’s own bookkeeping output from recursively triggering another edition, and the added test fixes that behavior explicitly. No content citation, assessment, confidential material, AGENTS.md amendment, or fabricated published record is introduced; the future-dated values appear only as synthetic test data.

The change updates `ledgerHash` in `src/domain/hash.ts` to exclude research agenda lifecycle bookkeeping fields (`status`, `statusNote`, `statusBy`, `statusDate`) so that an edition settling a research item does not self-trigger an immediate re-run on unchanged evidence. The modification is documented under `docs/DECISIONS.md` in accordance with §3.15 operator record rules (recorded from founder direction in session) and accompanied by unit tests in `src/domain/load.test.ts`. No constitutional rules, citation standards, or verification mechanisms are weakened or violated.

The diff only excludes research-agenda lifecycle fields (status, statusNote, statusBy, statusDate) from ledgerHash, mirrors the existing exclusion of source archival locators, and adds a regression test that settling an item leaves the hash unchanged. That is edition bookkeeping, not claim/evidence content, so it does not fabricate citations, hide assessments, edit AGENTS.md, or loosen §3.15 checks. The DECISIONS.md entry records a founder session direction with date and channel, which §3.15 permits. No content/ ledger records or runs are altered.

This is a targeted code fix, not a content or provenance change: the ledger hash stops counting the research agenda's lifecycle fields (status, statusNote, statusBy, statusDate), which the edition verb itself writes after its basis is taken. That exclusion does not loosen any check named in §3.15 — claims, evidence, sources, studies, and images remain fully hashed, so any change to records still triggers the scheduler, and nothing here bypasses CI, ratification, or parking; it only stops the edition's own bookkeeping from self-triggering a redundant edition, mirroring the existing exclusion of source locator fields. The settled status stays visible in the case history and agenda, so §3.14's visibility requirement is preserved, and a new test pins the behavior. The DECISIONS.md append is the workflow's required documentation (DECISIONS.md is not AGENTS.md, so §3.15's constitution-edit rule is not implicated), and the 'supervised-by' stamp is a normal dated provenance entry with nothing in the packet contradicting it. I see no fabrication, no confidence exposure, no misattribution, and no weakening of the fail-closed checks.