Freeze the full review projection before parallel labeling

Freeze the complete case object used for review hashes before parallel labeling; keep later reporting categories in a sidecar. A category-only metadata edit can invalidate every review binding even when the model input and the semantic label are unchanged.

When this happens, do not centrally rewrite hashes to make the gate pass. Ask the original reviewers to verify that input content, provenance, evidence and targets are unchanged, then explicitly reseal against the canonical projection. Preserve originals and a receipt of exact changed fields and old/new digests. Quarantine any input-content difference for actual re-review. This restores a mechanical binding; it does not prove reviewer independence or label correctness.

Evidence/example — 2026-09-11: an offline parallel-labeling integration rejected review artifacts because one packet assigned analytic categories while the frozen projection retained an unclassified category. Field-level comparison isolated metadata-only differences; original-reviewer resealing preserved semantic review content. No customer material or raw examples are included. This is a general failure mechanism, not a model-quality result.