The diff is infrastructure only: no content/ records, no citations added (verification section confirms none), and AGENTS.md is untouched. The new ClaimTreatment path lets an edition present a catalog claim's interpretation from the assessment it adopts while leaving the ledger record byte-identical (test asserts claims.yaml/claims-catalog.yaml bytes and loaded.claims unchanged), preserving statement, source anchor, origin, and independence group per §3.8/§3.10. Provenance and labeling are strengthened rather than weakened: the new ClaimAssessmentHistory shows model, runId, date, and 'AI-generated' vs 'human-reviewed' for every overlay, the claim page stamps adopted interpretations 'AI interpretation · date · standing', and unadopted drafts are explicitly described as not adopted (§3.15, §4 — no implying human review). The gating checks tighten rather than loosen: load.ts still fails closed when a selected claim lacks complete treatment, the schema requires a timestamped draft (not a 'check' role) and unique claim ids, the blind packet excludes treatment/essay/grades while adding selected catalog IDs to the grading scope, and evidencePacket now throws on multiple edition files instead of silently choosing one. Changing an interpretation alters the assessment hash and voids prior receipts, so fresh concurrence is required before standing returns (test asserts ratification drops to unratified, panel 0), consistent with §3.14/§3.15. The reviewCoverage narrowing to the current edition's selected claims tracks the displayed featured set and does not misstate human review. Nothing here fabricates records, exposes confidential material, or reclassifies a change to dodge the gate.

The change preserves catalog claims’ statements, source anchors, origins, independence groups, and ledger bytes while placing new interpretive fields in timestamped, hashed assessment overlays. Adopted interpretations are labeled with assessment date and standing, while unadopted records remain in an explicitly labeled assessment history with model, run ID, review status, and confidence. Validation fails closed when a selected catalog claim lacks a complete treatment, and the blind-review scope includes selected catalog claims without exposing the candidate interpretation or essay. The tests also show that revised interpretations lose prior standing, require fresh concurrence, and preserve earlier assessments; no citations, confidential material, AGENTS.md amendment, or concurrence-threshold weakening is introduced.

The change allows editions to feature catalog claims via versioned assessment treatments without mutating ledger records or source anchors, strictly adhering to §3.8 and §3.9. Clear separation between underlying propositions and AI interpretation is preserved and explicitly displayed. The new schema and snapshot utilities ensure that newly selected claims are included in independent blind review scopes and require fresh multi-model concurrence receipts if interpretation changes, upholding §3.14 and §3.15. No constitution files are altered and all tests demonstrate fail-closed validation.

The diff extends editions so a selected assessment can supply complete treatment (interpretation fields) for catalog claims without mutating ledger proposition, anchor, origin, tier, or independence group. Assessments stay append-only overlays; unadopted drafts are history-only; changed interpretation forces fresh concurrence; blind packets add selected IDs to grading scope while excluding treatment/essay/grades; multi-edition snapshots fail closed. No AGENTS.md edit, no citations added, no case judgments published—only synthetic fixtures and docs. UI keeps AI/draft/standing labels distinct from human ratification.

The change keeps catalog claims' ledger bytes immutable (tests assert claims-catalog.yaml is unchanged and statements/anchors/origin/independence groups are preserved), presents AI interpretations explicitly labeled 'AI interpretation' with date and standing, and marks history entries 'AI-generated' vs 'human-reviewed', satisfying §3.9, §3.14, and the no-human-provenance rule. Ratification integrity is preserved and arguably strengthened: selection alone does not mint standing (test asserts standing 'unratified'), fresh multi-seat concurrence receipts are required after any interpretation change (loss of standing after revised importance), blind packets include the new claims' propositions for independent grading while excluding treatment/prose/grades, and snapshots fail closed on ambiguous edition files rather than silently choosing one. The strict ClaimTreatmentSchema forbids ledger fields, preventing a treatment from replacing a proposition or fabricating provenance, and no citations are added. The validation change in load.ts still requires an assessment for every featured claim and complete treatment for catalog ones, so it does not weaken the checks (§3.15).