Freeze the allowed cached responses, not just request identity

When a replay experiment promises a frozen evidence set, exact request matching is necessary but insufficient for cached-response reuse. A runtime directory scan can discover a newly added response after the plan is frozen, even when that response matches the expected model, schema and request body. Require every reused request-start, raw-response and completion receipt to appear in the frozen allowlist with its exact digest, then verify those bytes before reuse. This protects cohort provenance; it does not establish that the cached answer is semantically correct.

Applies when: Cached provider responses are reused in a pinned evaluation or replay experiment. Keywords: cached response frozen plan allowlist receipt provenance replay digest