The diff is code-only (pipeline verification logic plus tests); it adds no content records, no citations, and the tooling found no identifiers to resolve. Both changes tighten rather than loosen the §3.15 checks: the ledger-anchor shortcut in verify.ts now applies only when a proposed claim's statement matches the ledger's wording verbatim (`anchoredByLedger`), so a rewritten claim reusing an existing id is refused as unanchored, and the reverify.ts cascade now leaves retired/superseded agenda items' status, claims and provenance untouched rather than rewriting them, which preserves provenance accuracy (§3.8, §3.14). New tests exercise both paths, including the refusal message and the byte-equality of the frozen agenda record. AGENTS.md is untouched, no confidential material appears, and nothing presents a draft as ratified. The only thing I would want the operator to confirm is that skipping the cascade for retired items cannot leave a retired item's claimIds pointing at a tombstoned claim and so break the fail-closed load; that is an ordinary engineering question, not a constitutional breach.

The change tightens the evidence-anchor shortcut by requiring a claim’s statement to match the ledger wording exactly, so evidence attached to one proposition cannot silently support rewritten words under the same ID. The added test confirms that such a rewritten claim is refused as unanchored. Leaving already retired or superseded research items unchanged preserves their recorded status and provenance rather than rewriting historical records during a later cascade. No constitution file, content citation, verification label, or ratification check is altered, and the tests cover both behavioral changes.

The change strengthens verification safeguards by ensuring that a re-read claim is only treated as anchored by existing ledger evidence if its text matches the ledger's statement verbatim, preventing semantic drift under reused IDs (§3.2, §3.15). It also prevents cascaded mutations on already retired or superseded research agenda items, preserving historical provenance and auditability (§3.14). Thorough test coverage is provided and no constitutional or content files are improperly touched.

The diff tightens claim anchoring so ledger live evidence only shortcuts a claim whose statement matches the ledger wording word-for-word; a rewritten claim on the same id is refused as unanchored. That closes the #401 loophole and upholds exact provenance and honest verification rather than weakening §3.15 checks. The cascade now skips research items already retired or superseded, so frozen status and provenance stamps are not rewritten. Tests cover both behaviors; no AGENTS.md edit, no content citations, and no confidence material are involved.

The change tightens rather than weakens the §3.15/§4 checks: previously any claim reusing a ledger id was anchored by the ledger's live evidence regardless of wording, so a rewritten claim could pass as evidence-anchored; the new `anchoredByLedger` requires an exact word-for-word statement match against the ledger's (or the settlement-supplied) wording, and a rewritten claim is now refused as unanchored — the opposite of check-weakening. The second change stops the retirement cascade from rewriting the status and provenance of already-retired or superseded agenda items, which serves §3.14's append-only, honestly-stamped history rather than misattributing who did what. Both behaviors are covered by added tests asserting refusal with a stated reason and byte-identical preservation of the frozen record. No citations are added, AGENTS.md is untouched, no runs are carried, and the diff is complete with no omitted files, so nothing hidden remains to judge.