{
  "workflow": "policy_grounding_review",
  "environment": "production",
  "issue_type": "refund",
  "customer_tier": "self-serve",
  "prompt_name": "refund-policy-check",
  "prompt_version": 4,
  "release_version": "2026.04.11",
  "model_provider": "openai",
  "model_name": "gpt-4o",
  "ticket_id": "ticket_damaged_item_policy_checker_invented_a_denial_rule_session_damage_claim_policy",
  "language": "en",
  "user_id": "customer_refund",
  "agent_profile_id": "policy-grounding-reviewer",
  "agent_team": "Trust and Safety",
  "agent_owner": "Policy quality",
  "agent_mission": "Checks whether policy-heavy responses remain grounded in approved support rules.",
  "story_label": "Damaged-item policy checker invented a denial rule",
  "story_summary": "The policy-grounding path fabricated unsupported policy text instead of asking for clarification or escalating.",
  "scenario_id": "scn_damaged_item_hallucination",
  "customer_intent": "Customer claimed the replacement item arrived broken and asked whether photo evidence was enough for a refund.",
  "expected_resolution": "Ask for clarification or escalate when the policy source is ambiguous, never invent denial language.",
  "policy_result": "hallucinated_policy_denial",
  "quality_score": 0.39,
  "severity_tag": "critical",
  "status_label": "Policy hallucination",
  "replay_label": "Policy Grounding Reviewer · Damaged-item policy checker invented a denial rule",
  "operator_notes": "Case: Damaged-item policy checker invented a denial rule\nIntent: Customer claimed the replacement item arrived broken and asked whether photo evidence was enough for a refund.\nExpected resolution: Ask for clarification or escalate when the policy source is ambiguous, never invent denial language.\nObserved policy result: hallucinated_policy_denial\nStory summary: The policy-grounding path fabricated unsupported policy text instead of asking for clarification or escalating.\nAgent role: Policy Grounding Reviewer\nOwner team: Trust and Safety"
}