Solutions
Clinical admissible automation (CTC)
CTC specifies permit‑to‑pay receipts for AI‑governed billing: replace narrative justification with replay‑verifiable evidence (including HL7 FHIR mappings).
Permit‑to‑payReplay‑verifiableAudit‑grade
Expected outputs: PASS (verifiable), FAIL (mismatch), HOLD (missing evidence/signatures; fail‑closed).
What can go wrong
- Denials, pend, or clawbacks because evidence is narrative (not replayable).
- Mismatch between clinical rationale, coding, and settlement‑critical actions.
- Model/policy drift changes outcomes without an auditable chain of custody.
- Disputes (appeals) lack a deterministic replay of what happened and why.
- Regulatory audits require provenance that teams cannot reproduce reliably.
What reviewers attach
- 1 URL (DSSE) —
/.well-known/mvg-procurement-ticket-pack.dsse.json. - 1 ZIP — Ticket Pack (offline verifier kit + sample packs).
- Security Review Packet (PDF) — architecture + threat model + evidence pointers.
- CTC references — CTC white paper (DOI 18340955) + CTC‑AAS constitution (PDF).
- Transparency — inclusion proof pointers for released receipts.
Where MVG gates
Verify receipts
Inputs, mappings, and policies are pinned; provenance is explicit and reviewable.
Permit‑to‑pay
Settlement‑critical actions require a permit backed by replayable evidence.
Admissible automation
If required receipts are missing or unverifiable, the workflow must HOLD for human review.
Standards & papers
Canonical pointers live on the standards surface; immutable deposits on Zenodo.