1 ROOT & GATE node §21.4 gate

Annex III FS Obligations Classifier art-238-classify-annex3-decisioning-obligations

EXTENDS run_ai_act_highrisk_fit (art-64). Maps Art 12(2) logging, Art 26(6) FRIA, and Art 27(1) DB registration obligations for Annex III 5(b) creditworthiness and 5(c) life/health insurance pricing AI. Gate: is_high_risk=false exits OUT_OF_SCOPE. Default (is_high_risk=true) proceeds to Art 12(2) record builder.

△ Gate: if /is_high_risk = false → END (OUT_OF_SCOPE) | default (true) → Step 2

MCP Call · classify_annex3_decisioning_obligations

{
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "name": "classify_annex3_decisioning_obligations",
    "arguments": {
      "is_high_risk": true,
      "annex3_category": "5b_creditworthiness",
      "deployer_role": "deployer",
      "has_human_oversight": true,
      "fria_completed": true,
      "db_registered": false,
      "logging_implemented": false
    }
  },
  "id": 1
}

2 D2 node

Art 12(2) AI Decision Log Record Builder art-236-build-ai-decision-log-record

Builds Art 12(2)-conformant decision log record: model_id, model_version, decision_label, confidence, input_digest (SHA-256), output_digest (SHA-256), sha256_prev_record (chain link), override_flag, override_by, subject_ref (structural only), retention_months (≥6), anchor_surface. Computes art12_completeness_score (0-12 fields). Passes sha256 record hash to IETF AAT validator.

MCP Call · build_ai_decision_log_record

{
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "name": "build_ai_decision_log_record",
    "arguments": {
      "model_id": "credit-score-v2",
      "model_version": "2.1.4",
      "decision_label": "APPROVED",
      "confidence": 0.91,
      "input_digest": "sha256:a3b4c5d6e7f8...",
      "output_digest": "sha256:b4c5d6e7f8a9...",
      "sha256_prev_record": "",
      "override_flag": false,
      "override_by": "",
      "subject_ref": "LOAN-REF-2026-001",
      "retention_months": 84
    }
  },
  "id": 2
}

3 TERMINAL node

IETF AAT Audit Trail Validator art-237-validate-agent-audit-trail

Validates IETF draft-sharif-agent-audit-trail-00 (expires 2026-09) compliance. Checks action_class (10 valid values), outcome (5 valid), trust_level (5 valid) enums; sha256_prev_record format (/^[0-9a-f]{64}$/); ECDSA signature presence. Emits conformance_result (CONFORMANT/PARTIAL/NON_CONFORMANT/EMPTY_INPUT) and aat_completeness_score (0-8). Terminal stage — exports composite §4 artifact.

MCP Call · validate_agent_audit_trail

{
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "name": "validate_agent_audit_trail",
    "arguments": {
      "agent_identity": "did:web:credit-model.example.com",
      "timestamp": "2026-07-04T12:00:00Z",
      "action_class": "AUTHORIZE",
      "outcome": "SUCCESS",
      "trust_level": "HIGH",
      "sha256_prev_record": "a3b4c5d6e7f8a9b0c1d2e3f4a5b6c7d8e9f0a1b2c3d4e5f6a7b8c9d0e1f2a3b4",
      "ecdsa_signature": "0x3045...",
      "reason": "Credit score 750 exceeds approval threshold 680"
    }
  },
  "id": 3
}

AI Decision Log Conformance