Self-inspecting, auditable memory for AI agents: it surfaces the evidence, provenance, and health behind each recall so callers can see which memory to trust. Optional Ed25519-signed tamper-evident le…