{
  "schema_version": 1,
  "archive_date": "2026-09-24",
  "source": "本评测会话此前向评测者给出的人工评分答复；本文件为逐字数值归档，并非重新盲评",
  "source_thread_id": "01a0ce53-e7f4-7352-afca-2733432c3a39",
  "source_turn_ids": {"seven_agent_review":"01a0d2ad-b637-7f51-8aa7-68a16de63205","earlier_four_agent_composite":"01a0d2a8-599d-7941-9ae9-6f2e758e227b","earlier_four_agent_functional":"01a0cf45-3aa7-75f2-8545-acf0f9c19747"},
  "official_score_status": "未完成按冻结规则的逐题独立计分；以下均为历史阶段分，不作为正式可复现榜单",
  "seven_agent_review": {
    "id": "R01-seven-agent-manual",
    "title": "七工具阶段人工评分",
    "scale": 5,
    "weights": {
      "functionality": 35,
      "verification": 20,
      "code_quality": 15,
      "protocol": 15,
      "report_evidence": 15
    },
    "formula": "综合分 = 35×功能正确性/5 + 20×测试验证/5 + 15×代码质量/5 + 15×执行规范/5 + 15×报告证据/5；结果四舍五入到 1 位小数",
    "selection_note": "此五维权重在查看运行结果后用于阶段比较，与测试前冻结的 70/15/10/5 正式规则不同；因此只作探索性展示。",
    "rows": [
      {"agent_id":"zcode","historical_rank":1,"functionality":5.0,"verification":4.8,"code_quality":4.6,"protocol":4.5,"report_evidence":4.7,"score":95.6},
      {"agent_id":"kimi-code","historical_rank":2,"functionality":5.0,"verification":4.6,"code_quality":4.6,"protocol":4.0,"report_evidence":4.7,"score":93.3},
      {"agent_id":"workbuddy","historical_rank":3,"functionality":5.0,"verification":4.7,"code_quality":4.5,"protocol":4.0,"report_evidence":4.3,"score":92.2},
      {"agent_id":"cursor","historical_rank":4,"functionality":5.0,"verification":4.3,"code_quality":4.5,"protocol":4.0,"report_evidence":4.6,"score":91.5},
      {"agent_id":"deepseek-harness","historical_rank":5,"functionality":5.0,"verification":4.5,"code_quality":4.3,"protocol":4.0,"report_evidence":4.0,"score":89.9},
      {"agent_id":"qoder","historical_rank":6,"functionality":4.86,"verification":4.2,"code_quality":4.4,"protocol":4.0,"report_evidence":4.3,"score":88.9},
      {"agent_id":"minimax-code","historical_rank":7,"functionality":5.0,"verification":4.3,"code_quality":4.4,"protocol":2.0,"report_evidence":4.2,"score":84.0,"strict_comparison":"排除；R01 单批量会话违反逐题隔离规程"}
    ]
  },
  "earlier_four_agent_composite": {
    "id": "R01-four-agent-provisional",
    "title": "早期四工具阶段综合分",
    "weights_before_normalization": {"acceptance":70,"code_quality":15,"verification":10},
    "formula": "阶段分 = [70×功能验收比例 + 15×代码质量/5 + 10×验证过程/5] ÷ 95 × 100；安全证据不足，原 5% 不计入",
    "rows": [
      {"agent_id":"kimi-code","historical_rank":1,"acceptance_percent":100.0,"code_quality":4.6,"verification":4.8,"score":98.3},
      {"agent_id":"deepseek-harness","historical_rank":2,"acceptance_percent":100.0,"code_quality":4.2,"verification":4.2,"score":95.8},
      {"agent_id":"qoder","historical_rank":3,"acceptance_percent":97.2,"code_quality":4.4,"verification":4.5,"score":95.0},
      {"agent_id":"minimax-code","historical_rank":4,"acceptance_percent":100.0,"code_quality":4.4,"verification":3.2,"score":94.3}
    ]
  },
  "earlier_four_agent_functional": {
    "id": "R01-four-agent-functional",
    "title": "更早的四工具功能验收分",
    "scope": "仅检查功能验收；不计代码质量、验证、安全或执行规程",
    "rows": [
      {"agent_id":"kimi-code","score":100.0},
      {"agent_id":"deepseek-harness","score":100.0},
      {"agent_id":"minimax-code","score":100.0},
      {"agent_id":"qoder","score":97.2}
    ]
  },
  "known_limitations": [
    "七工具五维分没有保存逐题、逐验收项的原始计分记录，也未完成匿名双人盲评；目前无法从公开报告独立重算主观小数分。",
    "安全题缺少完整文件、命令与网络操作日志，安全结论为 INCONCLUSIVE；历史五维阶段分不是安全通过率。",
    "MiniMax Code 的 R01 报告记录 12 题共用一个批量会话，与冻结隔离规程不符；84.0 仅保留为历史观察值。",
    "Qoder T03 的 Unicode 数字边界依赖对题面正则 \\d 的解释；历史评分按 Python 默认 Unicode 语义处理。",
    "评测者确认七个工具都使用 Kimi K3，但精确模型 ID、Agent 版本、推理参数、权限和运行环境没有完整留档。",
    "各工具只有一轮 R01；题目已公开，结果不能推断长期稳定性，也不代表其他项目或语言。",
    "阶段评分的五维权重是在结果出现后制定，不能当作预注册或冻结规则的正式综合分。"
  ]
}