Verify review artifacts before training-data access

A reviewed synthetic-data proposal should not become executable merely because release metadata says it is approved. Keep the new arm distinct, freeze prior builders and recipes, and require the training CLI to open and byte-verify an independent review receipt and supporting report before reading the corpus.

Bind raw and canonical proposal digests, parent dataset, exact retained/new counts and target distribution, independent author and reviewer identities, and any overlap audit. Adversarially mutate receipt bytes, content bindings, and approval fields and prove rejection occurs before data access.

This is an admission and provenance safeguard, not proof of label correctness, semantic independence, model improvement, deployment fidelity, or fresh validation. Results from a sealed archive remain bound to that archive.