Smallest possible proof that an AI action can be identity-bound, policy-checked, audited, and provenance-stamped.
No vendor integrations. The point is the receipt pipeline and what must be provable.
An auditor must be able to replay the chain and verify identity, policy, model, and provenance.
Every AI action can emit a structured receipt containing identity binding, token issuance, policy evaluation, audit events, and provenance stamps. This receipt is exportable and supports forensic replay.
This is a reference implementation. The real work is instrumenting your own AI systems with these primitives. Start with the Field Guide for implementation guidance.