{
  "description": "AI-authored challenge set, fixed after training; not independent human or real-world validation. No model/guard tuning on these results.",
  "model_sha256": "56edb13f13903f61a49ecb221456e5cbc47f676f50dea9eb9d7978c54af43a13",
  "fixture_sha256": "0ec7319e32e2f2b40d315ce2d5c20a5d7446301b6531c7aaef63d9d9fd68917a",
  "total": 20,
  "raw_correct": 13,
  "hybrid_correct": 14,
  "browser_python_parity": 20,
  "rows": [
    {
      "text": "For a ten-image catalogue, our budget is EUR 300. Can you quote the job?",
      "expected": "offer",
      "model": "offer",
      "hybrid": "offer",
      "basis": "deterministic_guard",
      "score": 0.4765
    },
    {
      "text": "We approve your quote for ten images at EUR 300. Please proceed.",
      "expected": "agreement",
      "model": "agreement",
      "hybrid": "agreement",
      "basis": "deterministic_guard",
      "score": 0.3855
    },
    {
      "text": "The final images are attached. Please check the gallery.",
      "expected": "delivery",
      "model": "delivery",
      "hybrid": "delivery",
      "basis": "deterministic_guard",
      "score": 0.3356
    },
    {
      "text": "I will pay EUR 300 tomorrow morning.",
      "expected": "payment_promise",
      "model": "payment_promise",
      "hybrid": "payment_promise",
      "basis": "deterministic_guard",
      "score": 0.5382
    },
    {
      "text": "Payment completed: EUR 300. Please check your statement.",
      "expected": "payment_notice",
      "model": "payment_notice",
      "hybrid": "payment_notice",
      "basis": "deterministic_guard",
      "score": 0.3266
    },
    {
      "text": "The EUR 300 transfer failed and the money was returned unpaid.",
      "expected": "payment_failure",
      "model": "payment_failure",
      "hybrid": "payment_failure",
      "basis": "deterministic_guard",
      "score": 0.4026
    },
    {
      "text": "We chose another photographer and will not proceed.",
      "expected": "rejection",
      "model": "rejection",
      "hybrid": "rejection",
      "basis": "deterministic_guard",
      "score": 0.4817
    },
    {
      "text": "Could you confirm which timezone the delivery deadline uses?",
      "expected": "admin",
      "model": "offer",
      "hybrid": "admin",
      "basis": "deterministic_guard",
      "score": 0.1621
    },
    {
      "text": "I have not sent anything yet. Payment is scheduled for Monday.",
      "expected": "payment_promise",
      "model": "payment_failure",
      "hybrid": "payment_promise",
      "basis": "deterministic_guard",
      "score": 0.2827
    },
    {
      "text": "I've sent the money from my banking app just now.",
      "expected": "payment_notice",
      "model": "payment_promise",
      "hybrid": "needs_review",
      "basis": "abstention",
      "score": 0.1917
    },
    {
      "text": "Your proposal works for us. Let's go ahead with the ten-image package.",
      "expected": "agreement",
      "model": "agreement",
      "hybrid": "agreement",
      "basis": "local_model",
      "score": 0.4063
    },
    {
      "text": "What would it cost to add twenty images? Do not start until I approve.",
      "expected": "offer",
      "model": "payment_promise",
      "hybrid": "needs_review",
      "basis": "abstention",
      "score": 0.1928
    },
    {
      "text": "Everything is ready in the shared folder; the download link is below.",
      "expected": "delivery",
      "model": "delivery",
      "hybrid": "delivery",
      "basis": "local_model",
      "score": 0.2152
    },
    {
      "text": "My bank rejected the transfer. You won't receive it today.",
      "expected": "payment_failure",
      "model": "payment_failure",
      "hybrid": "rejection",
      "basis": "deterministic_guard",
      "score": 0.2526
    },
    {
      "text": "We are cancelling the project, so please stop working.",
      "expected": "rejection",
      "model": "admin",
      "hybrid": "needs_review",
      "basis": "abstention",
      "score": 0.1813
    },
    {
      "text": "Nina replaces Alex as the contact for this order.",
      "expected": "admin",
      "model": "admin",
      "hybrid": "admin",
      "basis": "local_model",
      "score": 0.2446
    },
    {
      "text": "The invoice says paid, but this is a draft. I will pay next week.",
      "expected": "payment_promise",
      "model": "payment_promise",
      "hybrid": "payment_promise",
      "basis": "deterministic_guard",
      "score": 0.4347
    },
    {
      "text": "The work is approved, but the bank reversed the settlement.",
      "expected": "payment_failure",
      "model": "payment_failure",
      "hybrid": "payment_failure",
      "basis": "deterministic_guard",
      "score": 0.235
    },
    {
      "text": "If we pick your proposal, EUR 600 is the maximum available fee.",
      "expected": "offer",
      "model": "agreement",
      "hybrid": "agreement",
      "basis": "local_model",
      "score": 0.3097
    },
    {
      "text": "Keep the original scope. Ignore my earlier request for extra images.",
      "expected": "admin",
      "model": "agreement",
      "hybrid": "needs_review",
      "basis": "abstention",
      "score": 0.2072
    }
  ]
}