Repair digest-bound reviews through reviewer-owned canonical resealing

Digest-bound reviews should be repaired by re-binding the original reviewers’ unchanged semantics to the one canonical case projection, never by editing digests in place or letting a coordinator impersonate a reviewer. This applies when a mechanically irrelevant projection field (for example, a defaulted category) changes the case digest while the exact model input and provenance remain identical. Require an exact field-level diff, byte/canonical equality of every model-visible input and provenance field, per-review semantic digests before/after, independent resealing by each original reviewer, and quarantine on any broader difference. Then regenerate consensus from the canonical-bound reviews and verify that semantic statuses/targets are unchanged. Limits: this repairs cryptographic binding only; it does not strengthen reviewer independence, label quality, or correctness, and it is invalid if model input or substantive provenance changed.

Evidence/example — 2026-09-10: In an input-only tool-routing review, 60 reviews across two reviewers failed the canonical case binding because a category was explicitly set to a review-specific value instead of the canonical projection’s default. Each original reviewer independently resealed 30 reviews. Exact-difference receipts showed category was the sole difference; model input, input digest, source reference, provenance, call identity, node, and all semantic review fields were unchanged; zero cases were quarantined. Recomputed canonical consensus retained exactly 19 clear, 1 disagreement, and 10 unresolved cases, with 11 target-set disagreements. Provenance: deterministic local validation using training.reference_pilot.split.input_case and training.reference_pilot.review.review_case; original and repaired artifacts were content-hashed and originals preserved.