{
  "version": 1,
  "description": "Adversarial deterministic answer-quality proxy holdout. This validates metric behavior, not model quality.",
  "thresholds": {
    "minCases": 8,
    "minClassificationAccuracy": 1,
    "maxFalsePasses": 0
  },
  "cases": [
    {
      "id": "supported-git-policy",
      "shouldPass": true,
      "query": "How should I update protected main?",
      "answer": "Use a reviewed pull request and never force-push protected main [git-policy].",
      "referenceAnswer": "Use a reviewed pull request and never force-push protected main.",
      "contexts": [
        { "id": "git-policy", "text": "Use a reviewed pull request. Never force-push protected main." }
      ]
    },
    {
      "id": "negation-flip",
      "shouldPass": false,
      "query": "Can I force-push protected main?",
      "answer": "Force-push is safe on protected main [git-policy].",
      "referenceAnswer": "Never force-push protected main.",
      "contexts": [
        { "id": "git-policy", "text": "Never force-push protected main because it rewrites shared history." }
      ]
    },
    {
      "id": "numeric-drift",
      "shouldPass": false,
      "query": "Which build is live?",
      "answer": "Build 43 is live in production [health].",
      "referenceAnswer": "Build 42 is live in production.",
      "contexts": [
        { "id": "health", "text": "The production health response reports build 42." }
      ]
    },
    {
      "id": "off-topic-answer",
      "shouldPass": false,
      "query": "How do I prevent duplicate Stripe charges?",
      "answer": "Use medium heat and stir the sauce slowly [cooking].",
      "referenceAnswer": "Use a Stripe idempotency key.",
      "contexts": [
        { "id": "cooking", "text": "Use medium heat and stir the sauce slowly." }
      ]
    },
    {
      "id": "partial-hallucination",
      "shouldPass": false,
      "query": "What proves the deployment?",
      "answer": "The health endpoint matches build abc123 [deploy]. Revenue also doubled overnight.",
      "referenceAnswer": "The health endpoint must match build abc123.",
      "contexts": [
        { "id": "deploy", "text": "The production health endpoint matches build abc123." }
      ]
    },
    {
      "id": "invalid-citation",
      "shouldPass": false,
      "query": "What proves the deployment?",
      "answer": "The health endpoint matches build abc123 [invented-source].",
      "referenceAnswer": "The health endpoint must match build abc123.",
      "contexts": [
        { "id": "deploy", "text": "The production health endpoint matches build abc123." }
      ]
    },
    {
      "id": "supported-idempotency",
      "shouldPass": true,
      "query": "How do I prevent duplicate Stripe charges?",
      "answer": "Use the same Stripe idempotency key when retrying a PaymentIntent [billing-policy].",
      "referenceAnswer": "Use a Stripe idempotency key for PaymentIntent retries.",
      "contexts": [
        { "id": "billing-policy", "text": "Use the same Stripe idempotency key when retrying a PaymentIntent to prevent duplicate charges." }
      ]
    },
    {
      "id": "uncited-supported-claim",
      "shouldPass": false,
      "query": "How should I run a destructive migration?",
      "answer": "Create a recoverable database backup before the migration.",
      "referenceAnswer": "Create a recoverable database backup before a destructive migration.",
      "contexts": [
        { "id": "database-policy", "text": "Create a recoverable database backup before a destructive migration." }
      ]
    }
  ]
}