[
  {
    "id": "te05",
    "split": "test",
    "kind": "answerable",
    "question": "How should credentials be handled after exposure?",
    "expected_document": "d05",
    "expected_phrase": "revoke it",
    "result": {
      "status": "refused",
      "reason": "insufficient_lexical_evidence",
      "answer": null,
      "citations": [],
      "retrieved_ids": [
        "d05",
        "d01"
      ]
    },
    "correct": false,
    "gold_in_top3": true
  },
  {
    "id": "te08",
    "split": "test",
    "kind": "answerable",
    "question": "How long until the observations have to disappear?",
    "expected_document": "d01",
    "expected_phrase": "30 days",
    "result": {
      "status": "refused",
      "reason": "insufficient_lexical_evidence",
      "answer": null,
      "citations": [],
      "retrieved_ids": []
    },
    "correct": false,
    "gold_in_top3": false
  },
  {
    "id": "te11",
    "split": "test",
    "kind": "unanswerable",
    "question": "What GPU model does the compute budget require?",
    "result": {
      "status": "answered",
      "reason": "extractive_baseline",
      "answer": "The fictional teaching lab allows two GPU hours for each experiment. Stop the run at the budget limit and record both completed and failed trials.",
      "citations": [
        {
          "document_id": "d06",
          "title": "Compute budget",
          "quote": "The fictional teaching lab allows two GPU hours for each experiment. Stop the run at the budget limit and record both completed and failed trials."
        }
      ],
      "retrieved_ids": [
        "d06",
        "d03",
        "d02"
      ],
      "retrieval_score": 6.247025
    },
    "correct": false,
    "gold_in_top3": null
  },
  {
    "id": "te12",
    "split": "test",
    "kind": "unanswerable",
    "question": "Which bank reimburses the deployment budget?",
    "result": {
      "status": "answered",
      "reason": "extractive_baseline",
      "answer": "For a failed deployment in this fictional lab, restore the previous model artifact, run the health check, and record the rollback reason in the release log.",
      "citations": [
        {
          "document_id": "d03",
          "title": "Deployment rollback",
          "quote": "For a failed deployment in this fictional lab, restore the previous model artifact, run the health check, and record the rollback reason in the release log."
        }
      ],
      "retrieved_ids": [
        "d03",
        "d06"
      ],
      "retrieval_score": 2.678217
    },
    "correct": false,
    "gold_in_top3": null
  }
]
