[
 {
  "id": "T01",
  "answer_hash": "ebf692015b61",
  "correct": false,
  "reason": "答案建议直接复用旧 txt，并声称缓存向量不会漏结果，这与需在查询时重建文件列表且文件列表是结果白名单（遗漏文件即使有缓存向量也不会返回）的规则相矛盾。"
 },
 {
  "id": "T02",
  "answer_hash": "2da9346789db",
  "correct": false,
  "reason": "未说明 -format json 必须紧跟在 -monitor 之后，反而建议放在命令名后，且未指出放在命令末尾会导致 “Invalid parameter: -format” 错误。"
 },
 {
  "id": "T03",
  "answer_hash": "3f2b13b29fe0",
  "correct": false,
  "reason": "未按要求设置 git-builder: \"docker\"；答案用的是 build.type 而非明确的 git-builder 参数。"
 },
 {
  "id": "T04",
  "answer_hash": "3df043af2cf9",
  "correct": false,
  "reason": "未将 ClipboardSharingEnabled 设为 1，且没有重启 useractivityd（只涉及 pboard 与 sharingd），与要求不符。"
 },
 {
  "id": "T05",
  "answer_hash": "891e52eb93c3",
  "correct": false,
  "reason": "It fails to specify deleting tmp/openreview_token.json and that deleting it then retrying once resolves the 403, instead giving unrelated cache steps."
 },
 {
  "id": "T06",
  "answer_hash": "f7e8b0b75afd",
  "correct": false,
  "reason": "未按规则给出“每次间隔2-3秒”，反而建议5–8秒，与摘录不符。"
 },
 {
  "id": "T07",
  "answer_hash": "e4a9c02248bc",
  "correct": false,
  "reason": "It gives the 30‑minute interval but does not specify using ScheduleWakeup (nor the 1800s detail), so it fails to apply the required action from the rule."
 },
 {
  "id": "T08",
  "answer_hash": "f16c7c49eaf9",
  "correct": false,
  "reason": "答案错误地建议可直接用模型名作为 subagent_type；按规则 subagent_type 只能是已注册的 agent 名，否则会报 Unknown agent type，且需先在配置中把模型包装成一个 agent 再调用。"
 },
 {
  "id": "T09",
  "answer_hash": "eebf20de62e9",
  "correct": false,
  "reason": "应在配置顶层写 web_search = \"disabled\" 且值必须是字符串；答案写成 tools.web_search=false 并用布尔值，违反规则。"
 },
 {
  "id": "T10",
  "answer_hash": "d2ebcb0358fe",
  "correct": false,
  "reason": "答案建议在同一次运行中合并并去重，违背了规则“不能把归档与对应的原生库在同一次运行中同时使用，否则会把相同消息算两次”。"
 },
 {
  "id": "T11",
  "answer_hash": "e2013c328d1b",
  "correct": false,
  "reason": "未明确给出“每批最多50对、需拆分独立批次，且每批需重新审核与授权、不得复用令牌及不允许重试不确定批次”的规则。"
 },
 {
  "id": "T12",
  "answer_hash": "3c4a4545c475",
  "correct": false,
  "reason": "规则明确单个文件超过7200秒会在模型加载前被拒绝，因此2小时10分钟的单文件不能直接处理，答案与此相反。"
 },
 {
  "id": "T13",
  "answer_hash": "e70a2992e674",
  "correct": false,
  "reason": "未明确指出必须使用工作区内的相对路径，且未说明绝对路径会因 E_VIDEO_FILE 导致启动失败。"
 },
 {
  "id": "T14",
  "answer_hash": "9667b9e01b92",
  "correct": false,
  "reason": "未说明 40104 的根因是账号未完成验证，也未强调余额可用但关键词端点（含沙箱）被拦，且未给出在 DataForSEO 面板完成验证的处置。"
 },
 {
  "id": "T15",
  "answer_hash": "3b93f5fbf5eb",
  "correct": false,
  "reason": "未遵循 00:00–02:00 特例：凌晨 1:30 应用 --full-day 取昨天 00:00–今天 00:00，而不是今天 00:00–23:59；且全天在其他时段应到当前时刻。"
 },
 {
  "id": "T16",
  "answer_hash": "9454f976d2e4",
  "correct": false,
  "reason": "虽然说明了二选一，但它错误地建议用接近当前的显式时间戳且称不会被拒绝；正确做法是用 --departure-time now，且接近当前的时间戳可能会被 Routes API 拒绝。"
 },
 {
  "id": "T17",
  "answer_hash": "31dc19a4a95f",
  "correct": false,
  "reason": "不正确；结果不确定的提交必须阻止自动重提，且绝不能用新的 POST 去续上任务。"
 },
 {
  "id": "T18",
  "answer_hash": "4ade64b751aa",
  "correct": false,
  "reason": "规则要求样本必须在0–30秒内，超出的部分不会被裁剪而是直接拒绝作图，但你的回答称会原样画出并需手动截断，与规则相矛盾。"
 },
 {
  "id": "T19",
  "answer_hash": "6108d301b89d",
  "correct": false,
  "reason": "应说明 --tail 是布尔开关而非行数（不能用 --tail 100），并直接用 --tail 实时跟随日志；答案却声称没有 --tail 参数。"
 },
 {
  "id": "T20",
  "answer_hash": "0bb97e6ccc49",
  "correct": false,
  "reason": "未明确指出“开始清扫仅在机器人停靠底座时有效，不在底座就会返回NOT_READY”，而是给出多种等权原因的排查。"
 },
 {
  "id": "T21",
  "answer_hash": "d6bf7fb7d070",
  "correct": false,
  "reason": "答案未说明需使用带末尾8位日期的 canonical_slug，且声称用 author/model-name 即可，与“短 slug 会返回空数组、需传 canonical_slug”的规则相矛盾。"
 },
 {
  "id": "T22",
  "answer_hash": "6aa738fda25a",
  "correct": false,
  "reason": "默认包含子会话说对了，但应使用 --no-expand 禁止扩展，答案给了错误的 --no-recursive。"
 },
 {
  "id": "T23",
  "answer_hash": "bfd9aa8fcbcb",
  "correct": false,
  "reason": "虽然指出先下混会导致假削波，但未提及相关立体声约 +3 dB 的误差来源，且未提供 ebur128=peak=true 的交叉验证命令。"
 },
 {
  "id": "T24",
  "answer_hash": "5b894b7502fd",
  "correct": false,
  "reason": "未给出12–20词的建议范围，且错误声称没有硬性上限并建议可达25词，违背“普通句子不得超过22词”的硬性上限。"
 },
 {
  "id": "T25",
  "answer_hash": "167fb49b0010",
  "correct": false,
  "reason": "未说明对 search/extract 加 --stdout 会报未识别参数并以退出码 2 退出且不发起请求，且错误声称默认会把完整结果打印到 stdout 并正常发起请求。"
 },
 {
  "id": "T26",
  "answer_hash": "66a87471d9c1",
  "correct": false,
  "reason": "答案将强拍改动放宽到每小节一次（8/8），未要求记录在 skeleton_allow，且误解强拍应为每小节的第1与第3拍，违背“每四小节最多一次且需记载，通常只在终止式”的规则。"
 },
 {
  "id": "T27",
  "answer_hash": "dc72457256fb",
  "correct": false,
  "reason": "说明了在 Gemini 下会被忽略，但没有指出该参数仅对 GPT Image 模型生效。"
 },
 {
  "id": "T28",
  "answer_hash": "f1452d0308f6",
  "correct": false,
  "reason": "未按规则处理：未提到在 409 时用同样请求并带 \"confirm_conflict\": true 以并存，也未说明先用 POST /scheduled/{id}/cancel 取消旧任务再替换；转而讨论 cron/Windows 不符合给定规则。"
 },
 {
  "id": "T29",
  "answer_hash": "ce32bd884cd4",
  "correct": false,
  "reason": "答案错误：--after/--before 不是时间过滤器，不能传 ISO 时间；它们需要 UUID 游标，正确做法是先导出最近邮件再按 created_at 本地过滤。"
 },
 {
  "id": "T30",
  "answer_hash": "405dd2fab8c8",
  "correct": false,
  "reason": "未按规则使用 Typefully 长文为默认且未说明“单条推文”指一条 Typefully 帖子，并把切到短推的参数错写为 tweet_mode/截断而非 --no-long-post。"
 }
]