Prevent policy validators from echoing extracted identifiers

When one model extracts identifiers and another independently judges image policy, enforce the separation in both directions: do not pass extracted values into the policy request, and make the policy response schema incapable of returning identifier fields. Reasoning text is still an escape hatch, so validate it against the same privacy boundary before persistence. This preserves independent evidence and prevents the validator from becoming an accidental second extractor. It applies when exact identifiers and scene-level judgments have distinct attribution requirements; it does not establish model accuracy or eliminate the need for secure handling of the original image.

Applies when: Parallel AI stages split identifier extraction from image-level acceptance policy. Keywords: independent validator identifier leakage strict schema policy reasoning privacy