{
  "created_at": "2026-09-25T18:49:48.190385+00:00",
  "model": "gpt-6-luna",
  "reasoning_effort": "low",
  "corpus_sha256": "6da6c090c762533f857738a8f629c7c56ca12acabf9a034a57672bf8817771fd",
  "subject_passes": 1,
  "total": 3,
  "task_errors": 0,
  "notes": "Three constructed first messages, real T3 title generation through the production prompt builder and Codex adapter. No prior title or later messages. The demo branch deliberately reproduces a truncation regression. Subject checks use vocabulary groups; format counts are observations. This is not a model benchmark or an agent repair success-rate measurement.",
  "cases": [
    {
      "case_id": "initial-offline-search",
      "title": "Fix Stale Label Palette Cache",
      "checks": {
        "identifies_subject": false
      },
      "format_observations": {
        "characters": 29,
        "words": 5
      }
    },
    {
      "case_id": "initial-draft-recovery",
      "title": "Fix Stale Connection Badge Status",
      "checks": {
        "identifies_subject": false
      },
      "format_observations": {
        "characters": 33,
        "words": 5
      }
    },
    {
      "case_id": "initial-checkout",
      "title": "Prevent Duplicate Checkout Charges",
      "checks": {
        "identifies_subject": true
      },
      "format_observations": {
        "characters": 34,
        "words": 4
      }
    }
  ]
}