{
  "status": "complete",
  "detail": "冻结48案例各3次，完整真实模型回归复测；140通过、4失败均公开。",
  "updated_at": "2026-10-21",
  "evaluation": {
    "status": "complete",
    "passed": 140,
    "total": 144,
    "rate": 97.2,
    "cases": 48,
    "repetitions": 3,
    "concurrency": 2,
    "report_url": "https://github.com/yaowcui-ai/aftersales-agent/blob/main/evaluation/results/latest.json"
  },
  "tests": {
    "passed": 80,
    "detail": "离线应用测试通过，使用替身模型与隔离数据库"
  },
  "browser": {
    "steps": 7,
    "detail": "真实浏览器：新对话、追问、历史切换和刷新恢复"
  },
  "evidence": [
    {
      "label": "逐次 JSONL",
      "url": "https://github.com/yaowcui-ai/aftersales-agent/blob/main/evaluation/results/latest.jsonl"
    },
    {
      "label": "SHA256 与失败索引",
      "url": "https://github.com/yaowcui-ai/aftersales-agent/blob/main/evaluation/results/manifest.json"
    },
    {
      "label": "历史 129/144 原始结果",
      "url": "https://github.com/yaowcui-ai/aftersales-agent/blob/main/evaluation/results/historical-20261020.json"
    }
  ]
}