{
  "schemaVersion": "1.2.0",
  "default_evaluators": {
    "Relevance": {},
    "Coherence": {}
  },
  "items": [
    {
      "name": "Expense Policy Flow",
      "description": "Test that agent handles expense policy questions across turns",
      "turns": [
        {
          "prompt": "I'm traveling to Seattle next week for a conference.",
          "expected_response": "I can help with travel-related questions."
        },
        {
          "prompt": "My dinner last night was $250. Is that okay?",
          "expected_response": "The per-diem meal allowance is a maximum of $200.",
          "evaluators": {
            "Groundedness": { "threshold": 4 }
          }
        },
        {
          "prompt": "What should I do about the overage?",
          "expected_response": "For expenses exceeding the policy limit, you'll need manager approval.",
          "evaluators": {
            "ExactMatch": { "case_sensitive": false }
          },
          "evaluators_mode": "replace"
        }
      ]
    }
  ]
}