{
  "schema_version": "preflight/v2",
  "base_url": "http://127.0.0.1:8000/v1",
  "model": "vision.general",
  "thinking": {
    "mode": "disabled",
    "chat_template_kwargs": {
      "enable_thinking": false
    },
    "reasoning_effort": null
  },
  "budget": {
    "visible_answer_tokens": 1024,
    "reasoning_headroom_tokens": 0,
    "max_completion_tokens": 1024
  },
  "multimodal_input": {
    "bytes": 14307,
    "mime": "image/png",
    "sha256": "c104d086a52a0b60981f6490074b476f20d2a45683900fa372c5edeaff0fab51"
  },
  "image_expect": [
    "RTX 5090",
    "503",
    "Retry"
  ],
  "ocr_expect": [],
  "checks": [
    "image"
  ],
  "results": [
    {
      "name": "general image understanding",
      "passed": true,
      "detail": "4.6s all expected text present finish='stop' visible=1025 reasoning_chars=0 reasoning_tokens=None"
    }
  ],
  "observations": [
    {
      "content": "\nThe image shows the **Anvil Serving Dashboard**, which provides system status and request metrics. Key details include:\n\n### **GPU Status (Left Panel)**\n- **GPU**: RTX 5090 \n- **VRAM used**: 30057 MiB \n- **Serves running**: 7 \n- **Status**: **HEALTHY** (indicated by a green dot). \n\n### **Error Notification (Right Panel)**\n- **Error Code**: 503 \n- **Message**: *\"No available tier for preset 'planning' - all candidates busy\"* \n- **Action**: A blue **\"Retry\"** button is available. \n\n### **Requests per Tier (Bottom Chart)**\n- A bar chart titled **\"Requests per tier\"** shows three tiers: \n - **heavy** (blue bar): Highest request volume. \n - **fast** (orange bar): Moderate request volume. \n - **ocr** (green bar): Lowest request volume. \n\n### **Summary of Critical Information**\n- The system is **healthy** (GPU operational), but there is a **503 Service Unavailable error** due to all candidates being busy for the `'planning'` tier. \n- Request distribution is uneven, with the **heavy** tier handling the most requests.",
      "finish_reason": "stop",
      "content_chars": 1025,
      "content_excerpt": "\nThe image shows the **Anvil Serving Dashboard**, which provides system status and request metrics. Key details include:\n\n### **GPU Status (Left Panel)**\n- **GPU**: RTX 5090 \n- **VRAM used**: 30057 Mi",
      "reasoning_field": null,
      "reasoning_chars": 0,
      "reasoning_excerpt": "",
      "reasoning_tokens": null,
      "usage": {
        "prompt_tokens": 562,
        "completion_tokens": 622,
        "total_tokens": 1184
      },
      "test": "image",
      "seconds": 4.589,
      "passed": true,
      "validation_detail": "all expected text present",
      "image": {
        "bytes": 14307,
        "mime": "image/png",
        "sha256": "c104d086a52a0b60981f6490074b476f20d2a45683900fa372c5edeaff0fab51"
      },
      "expected": [
        "RTX 5090",
        "503",
        "Retry"
      ]
    }
  ],
  "evidence_policy": {
    "reasoning": "any",
    "allowed_finish_reasons": [
      "stop",
      "tool_calls"
    ],
    "errors": []
  },
  "passed": true
}