Skip to content
AIGP

RFC-033: Quantitative Outcome Evaluation Model — 5. Protocol Integration

AIGP Specification › RFC-033: Quantitative Outcome Evaluation Model › 5. Protocol Integration

← 4. Temporal Model · Section index · 6. Evaluation Pipeline →

5. Protocol Integration

5.1 RECORD Extension

Each RECORD message gains an optional outcome_score field:

{
  "message_type": "RECORD",
  "app_id": "SCIM_PRJ",
  "use_case": "compliance_query",
  "status": "SUCCESS",
  "outcome_score": {
    "composite": 0.84,
    "dimensions": {
      "correctness": 0.88,
      "completeness": 0.82,
      "relevance": 0.90,
      "safety": 1.00,
      "efficiency": 0.65,
      "consistency": 0.78
    },
    "band": "GOOD",
    "baseline_delta": +0.02,
    "evaluated_by": "bedrock:us.anthropic.claude-haiku-4-5",
    "evaluated_at": "2026-06-25T01:00:00Z"
  }
}

5.2 Contract Integration (AR Rules)

ar_rules:
  - id: AR-QUALITY-01
    description: "Minimum outcome score for autonomous operation"
    condition: outcome_score.composite < 0.60
    action: RESTRICT_SCOPE


  - id: AR-QUALITY-02
    description: "Quality regression detection"
    condition: outcome_score.baseline_delta < -0.15
    action: ALERT_AND_REVIEW


  - id: AR-QUALITY-03
    description: "Earned autonomy — consistently high quality"
    condition: rolling_mean(outcome_score.composite, 7d) > 0.90
    action: EXPAND_AUTONOMY

5.3 VERIFY Integration (RFC-032)

The VERIFY verdict incorporates the Outcome Score:

{
  "message_type": "VERIFY",
  "verdict": "MATCH",
  "outcome_score": 0.84,
  "anticipation_alignment": 0.91,
  "combined_confidence": 0.87
}

Where:

  • outcome_score = quality of what was produced
  • anticipation_alignment = how well it matched what was predicted
  • combined_confidence = weighted merge of both

← 4. Temporal Model · Section index · 6. Evaluation Pipeline →