Runtime Alignment Monitor Data Contracts

Runtime Analyze Input

Field	Type	Required	Description
agent_id	string	yes	Monitored agent identity.
objective_context	object	yes	Principal intent and policy boundaries.
events	object[]	yes	Runtime events used for scoring.
interventions	object[]	no	Human intervention observations.

Objective Context Example

{
  "principal": "compliance-team",
  "allowed_objectives": ["contract-fulfillment", "safety"],
  "forbidden_objectives": ["hidden-commission-maximization"],
  "policy_version": "v3.2"
}

Recommendation Output Schema

{
  "scores": {
    "drift": 38.4,
    "fidelity": 61.9,
    "mandate": 52.7,
    "meridian": 70.1
  },
  "confidence": 0.83,
  "recommendation": "review",
  "risk_band": "high",
  "evaluated_at": "2026-03-04T14:00:00Z"
}

Webhook Event Shape (runtime.risk.elevated)

{
  "event": "runtime.risk.elevated",
  "event_id": "evt_3001",
  "agent_id": "agent_42",
  "recommendation": "degrade",
  "risk_band": "critical",
  "scores": {
    "drift": 19.0,
    "fidelity": 48.1,
    "mandate": 44.5,
    "meridian": 68.9
  }
}