Metered AI governance for
Medicaid managed care utilization review

A multi-state government-program health plan

This demo models prior authorization support for a Medicaid-first health plan operating across state programs, provider networks, clinical review queues, appeals, audit requests, and federal reporting requirements.

Budgets are controlled at the client, line-of-business, workflow, case, and agent level. Receipts prove individual operations.

1. 01Packet intake classification
2. 02Document extraction
3. 03Completeness check
4. 04Criteria mapping
5. 05Reviewer summary generation
6. 06RFI reason draft
7. 07Escalation recommendation
8. 08Evidence bundle export

Each step is an AI-assisted draft for human review. No step is a coverage decision.

Each receipt records one metered AI action. The receipt is evidence, not the budget.

CMS Prior Authorization Readiness

This demo models workflow readiness for CMS prior authorization timing, reason specificity, status traceability, and future FHIR prior authorization API integration. It does not submit a real prior authorization request.

Export contains synthetic case identifiers, operation receipts, budget hierarchy state, documentation completeness, criteria mapping, draft reason, and a compliance trace. It does not contain real PHI.

{
  "packetType": "synthetic_prior_authorization_oversight_packet",
  "caseId": "SYN-CASE-BH-2026-041",
  "tenantId": "tenant_government_program_payer",
  "lineOfBusiness": "medicaid_mco",
  "programProfile": "State A Medicaid Program",
  "requestType": "Behavioral Health Inpatient Extension",
  "urgency": "expedited",
  "decisionClock": "72 hours",
  "receivedAt": "2026-05-27T09:00:00Z",
  "aiReviewStartedAt": "2026-06-20T15:35:03.384Z",
  "humanReviewRequired": true,
  "humanDecision": null,
  "aiOperations": [],
  "budgetHierarchy": {
    "tenantId": "tenant_government_program_payer",
    "clientMonthlyBudgetUsd": 50000,
    "lineOfBusinessBudgetUsd": 20000,
    "workflowBudgetUsd": 8000,
    "caseCapUsd": 0.15,
    "agentCapsUsd": {
      "documentation-extraction-agent": 0.03,
      "completeness-check-agent": 0.02,
      "criteria-mapping-agent": 0.04,
      "reviewer-summary-agent": 0.05,
      "rfi-reason-agent": 0.02,
      "escalation-recommendation-agent": 0.02,
      "evidence-export-agent": 0.03
    },
    "currentSpendUsd": {
      "client": 0,
      "lineOfBusiness": 0,
      "workflow": 0,
      "case": 0,
      "agents": {}
    }
  },
  "documentationCompleteness": {
    "Admission note": "complete",
    "Treatment plan": "complete",
    "Risk assessment": "partial",
    "Discharge plan": "missing",
    "Medication reconciliation": "missing",
    "Current symptom update": "partial"
  },
  "criteriaMapping": [
    {
      "category": "Acute safety risk",
      "extractedEvidence": "Provider notes ongoing acute symptoms with active safety monitoring required.",
      "status": "reviewer_required",
      "confidence": 0.62,
      "reviewerNote": "Risk assessment is partial; reviewer must validate current safety status."
    },
    {
      "category": "Current symptoms",
      "extractedEvidence": "Current symptom update is partial. Some fields missing.",
      "status": "not_enough_information",
      "confidence": 0.48,
      "reviewerNote": "Request more information from provider on current symptom inventory."
    },
    {
      "category": "Treatment response",
      "extractedEvidence": "Treatment plan documents medication and therapy regimen in place.",
      "status": "met",
      "confidence": 0.81,
      "reviewerNote": "Treatment plan present and consistent."
    },
    {
      "category": "Discharge readiness",
      "extractedEvidence": "Discharge plan is incomplete.",
      "status": "not_enough_information",
      "confidence": 0.35,
      "reviewerNote": "A complete discharge plan is required before evaluating discharge readiness."
    },
    {
      "category": "Less restrictive level of care",
      "extractedEvidence": "No assessment of step-down alternatives is included.",
      "status": "not_enough_information",
      "confidence": 0.4,
      "reviewerNote": "Request step-down alternatives assessment."
    },
    {
      "category": "Medication plan",
      "extractedEvidence": "Medication reconciliation is missing.",
      "status": "not_enough_information",
      "confidence": 0.3,
      "reviewerNote": "Medication reconciliation must be supplied."
    },
    {
      "category": "Follow-up plan",
      "extractedEvidence": "Outpatient follow-up plan is referenced but not confirmed.",
      "status": "reviewer_required",
      "confidence": 0.55,
      "reviewerNote": "Reviewer to validate outpatient follow-up scheduling."
    }
  ],
  "draftReason": "The packet is missing a complete discharge plan and current medication reconciliation. A behavioral health risk assessment is partial and requires reviewer validation.",
  "complianceTrace": {
    "cmsDecisionClockTracked": true,
    "specificReasonGenerated": true,
    "humanReviewBoundaryPreserved": true,
    "syntheticDataOnly": true,
    "realPhiProcessed": false
  },
  "evidenceHash": "ev_SYN_PACKET_-041",
  "exportedAt": "2026-06-20T15:35:03.384Z"
}

Compliance Trace

• CMS decision-clock trackedpending
  Each AI operation carries a timestamp.
• Specific reason generatedpending
  Structured RFI / adverse-reason draft attached.
• Human review boundary preservedok
  No autonomous denial path exists.
• Synthetic data onlyok
  No real PHI processed.
• Human decision recordedpending
  Pending reviewer action

How P402 backs every receipt

Each operation receipt is one metered AI action settled through P402's payment-aware orchestration layer: x402 for per-operation payment authorization (EIP-3009 USDC.e on Tempo mainnet), AP2 mandates for client-level spending authority, policy enforcement at the tenant / line-of-business / workflow / case / agent level, and evidence bundles that bind operation, cost, model, policy decision, and reviewer outcome.

Receipts are evidence. Budgets live in policy. Decisions live with the reviewer.