{
  "schema_version": "preflight/v2",
  "phase": "incumbent_after_restoration",
  "base_url": "http://127.0.0.1:8000/v1",
  "model": "llm.primary",
  "budget": {
    "max_completion_tokens": 512,
    "reasoning_headroom_tokens": 0,
    "visible_answer_tokens": 512
  },
  "checks": ["json"],
  "thinking": {
    "mode": "disabled",
    "reasoning_effort": null,
    "chat_template_kwargs": {"enable_thinking": false}
  },
  "sampling": {"temperature": 0.0},
  "passed": false,
  "observations": [
    {
      "test": "json",
      "content": "The user wants me to return ONLY a JSON object: {\"language\":\"python\",\"ok\":true}. No prose.\n\nThis is a simple, clear instruction. I should just output the JSON object exactly as requested, with no additional text.{\"language\":\"python\",\"ok\":true}",
      "content_chars": 243,
      "finish_reason": "stop",
      "reasoning_chars": 0,
      "reasoning_field": null,
      "usage": {
        "prompt_tokens": 30,
        "completion_tokens": 58,
        "total_tokens": 88,
        "reasoning_tokens": 0,
        "cached_tokens": 0
      }
    }
  ],
  "results": [
    {
      "name": "structured JSON",
      "passed": false,
      "detail": "error: Extra data: line 1 column 32 (char 31)"
    }
  ],
  "projection": {
    "native_source_sha256": "24340ec4aca45a8ca83ce1f0ba85335c8df9a122f68051ed4da9ba7e718c100b",
    "transformations": [
      "Removed elapsed timing because this is not a performance comparison.",
      "Preserved the complete visible response, token usage, controls and strict JSON failure."
    ]
  }
}