{
  "rows": [
    {
      "benchmark": "MMTB",
      "case": "三路机位 → 跟随当前发言人剪辑",
      "case_id": "mmtb-multicam",
      "source_kind": "本地实测 · 不限时",
      "model": "gpt-5.6-sol",
      "harness": "Codex 0.146.0",
      "responses": 45,
      "outer_tools": 44,
      "expanded_tools": 44,
      "wait": 25,
      "nonwait": 19,
      "score": 0.8333,
      "passed": false,
      "note": "含中断前与恢复后两段；未重复计入导入历史",
      "agent_config": "GPT-5.6 Sol + Codex 0.146.0"
    },
    {
      "benchmark": "MMTB",
      "case": "对照乐谱 → 找出钢琴演奏错误",
      "case_id": "mmtb-piano",
      "source_kind": "本地实测 · 不限时",
      "model": "gpt-5.6-sol",
      "harness": "Codex 0.146.0",
      "responses": 31,
      "outer_tools": 30,
      "expanded_tools": 32,
      "wait": 22,
      "nonwait": 10,
      "score": 0.8,
      "passed": true,
      "note": "单次独立运行",
      "agent_config": "GPT-5.6 Sol + Codex 0.146.0"
    },
    {
      "benchmark": "MMTB",
      "case": "听到的字段 ≠ 焦点字段？无障碍审查",
      "case_id": "mmtb-accessibility",
      "source_kind": "本地实测 · 不限时",
      "model": "gpt-5.6-sol",
      "harness": "Codex 0.146.0",
      "responses": 22,
      "outer_tools": 21,
      "expanded_tools": 28,
      "wait": 12,
      "nonwait": 16,
      "score": 1.0,
      "passed": true,
      "note": "单次独立运行",
      "agent_config": "GPT-5.6 Sol + Codex 0.146.0"
    },
    {
      "benchmark": "ALE",
      "case": "镜头日志",
      "case_id": "ale-storyboard",
      "source_kind": "官方公开轨迹",
      "model": "claude-opus-4-8",
      "harness": "claude_code",
      "responses": null,
      "outer_tools": 53,
      "expanded_tools": 53,
      "wait": 0,
      "nonwait": 53,
      "score": 1.0,
      "passed": null,
      "note": "API 响应边界无法可靠还原",
      "agent_config": "Claude Opus 4.8 + Claude Code"
    },
    {
      "benchmark": "ALE",
      "case": "音乐转谱",
      "case_id": "ale-music",
      "source_kind": "官方公开轨迹",
      "model": "anthropic/claude-opus-4.7",
      "harness": "openclaw",
      "responses": null,
      "outer_tools": 222,
      "expanded_tools": 222,
      "wait": 61,
      "nonwait": 161,
      "score": 0.2220804767923586,
      "passed": null,
      "note": "API 响应边界无法可靠还原",
      "agent_config": "Claude Opus 4.7 + OpenClaw"
    },
    {
      "benchmark": "ALE",
      "case": "白鹤抠像",
      "case_id": "ale-chroma",
      "source_kind": "官方公开轨迹",
      "model": "claude-opus-4-8",
      "harness": "claude_code",
      "responses": null,
      "outer_tools": 197,
      "expanded_tools": 197,
      "wait": 7,
      "nonwait": 190,
      "score": 0.5674291278881924,
      "passed": null,
      "note": "API 响应边界无法可靠还原",
      "agent_config": "Claude Opus 4.8 + Claude Code"
    }
  ],
  "means": {
    "MMTB": {
      "responses": 32.6667,
      "outer_tools": 31.6667,
      "expanded_tools": 34.6667,
      "wait": 19.6667,
      "nonwait": 15.0,
      "score": 0.8777666666666667,
      "agent_config": "GPT-5.6 Sol + Codex 0.146.0"
    },
    "ALE": {
      "responses": null,
      "outer_tools": 157.3333,
      "expanded_tools": 157.3333,
      "wait": 22.6667,
      "nonwait": 134.6667,
      "score": 0.5965032015601837,
      "agent_config": "混合配置（见上）"
    }
  },
  "checked_on": "2026-09-23"
}