[
 {
  "id": "T01",
  "answer_hash": "afad66749b01",
  "correct": true,
  "reason": "答案明确说明查询时需按当前范围重建文件列表且不能复用旧列表，并指出文件列表是结果白名单、遗漏文件即使缓存有向量也不会返回。"
 },
 {
  "id": "T02",
  "answer_hash": "b93b9cec8081",
  "correct": true,
  "reason": "答案明确要求将 -format json 紧跟在 -monitor 之后，并指出放在命令末尾会触发 Invalid parameter: -format 错误。"
 },
 {
  "id": "T03",
  "answer_hash": "39c64414cb8a",
  "correct": true,
  "reason": "It explicitly instructs to set the git-builder input to \"docker,\" which is the required flag to force Docker builds instead of buildpack."
 },
 {
  "id": "T04",
  "answer_hash": "e15818e3f0ff",
  "correct": true,
  "reason": "It sets ClipboardSharingEnabled=1 in ~/Library/Preferences/com.apple.coreservices.useractivityd.plist and restarts useractivityd, sharingd, and pboard as required."
 },
 {
  "id": "T05",
  "answer_hash": "d26df813c045",
  "correct": true,
  "reason": "明确指出删除 tmp/openreview_token.json 并且只需重试一次即可，且未与规则矛盾。"
 },
 {
  "id": "T06",
  "answer_hash": "995222b96838",
  "correct": true,
  "reason": "答案明确指出每次下载之间应间隔 2-3 秒，与规则一致且无矛盾。"
 },
 {
  "id": "T07",
  "answer_hash": "cdaac06f5e60",
  "correct": true,
  "reason": "明确写出默认约30分钟（1800s）的巡检间隔，并说明用 ScheduleWakeup 安排唤醒，且未与规则矛盾。"
 },
 {
  "id": "T08",
  "answer_hash": "e157f1ccb05c",
  "correct": true,
  "reason": "答案明确说明 subagent_type 只能填已注册的 agent 名、填模型名会报 Unknown agent type，并给出需在配置中先将模型包装成 agent 再调用的做法。"
 },
 {
  "id": "T09",
  "answer_hash": "e48144c42d69",
  "correct": true,
  "reason": "答案明确说明需在基础配置或所加载的 profile 的配置顶层写 web_search = \"disabled\"，且值只能是字符串 \"disabled\"（布尔等都会被拒绝），与摘录完全一致。"
 },
 {
  "id": "T10",
  "answer_hash": "bd7b501de6e8",
  "correct": true,
  "reason": "答案明确说明不能在同一次运行中混用归档与对应的原生库，理由是会把相同消息算两次。"
 },
 {
  "id": "T11",
  "answer_hash": "ef239aa28570",
  "correct": true,
  "reason": "答案明确写出每批最多50对且需拆成独立批次、每批需重新审核与授权、令牌不可复用并且不允许重试不确定批次，且未与规则矛盾。"
 },
 {
  "id": "T12",
  "answer_hash": "14013261935f",
  "correct": false,
  "reason": "答案开头声称“2小时10分可以处理、未超限”，与规则“单个文件超7200秒会在加载模型前被拒绝”相矛盾，尽管随后更正但已出现矛盾。"
 },
 {
  "id": "T13",
  "answer_hash": "3d3f32982458",
  "correct": true,
  "reason": "答案明确说明视频文件必须用工作区内的相对路径，且使用绝对路径会以 E_VIDEO_FILE 启动失败，与摘录一致。"
 },
 {
  "id": "T14",
  "answer_hash": "1a7b516d18e2",
  "correct": true,
  "reason": "答案明确指出 40104 源于账号未完成验证，说明余额可用但关键词端点（含沙箱）被拦截，并给出在 DataForSEO 面板完成验证的解决方案。"
 },
 {
  "id": "T15",
  "answer_hash": "50047f9e4559",
  "correct": false,
  "reason": "虽然说明了无需手算并给出 00:00–02:00 取昨全天，但未明确说明其他时间段应取“今天 00:00 到当前时刻”。"
 },
 {
  "id": "T16",
  "answer_hash": "f9ee4230d2ad",
  "correct": true,
  "reason": "明确说明只能二选一，并指出“现在出发”应使用 --departure-time now，且不要传接近当前的显式时间戳以免被 Routes API 拒绝。"
 },
 {
  "id": "T17",
  "answer_hash": "1d5df8732ca4",
  "correct": true,
  "reason": "答案明确指出结果不明时不得再次提交，并强调必须阻塞自动重提且不要用新的 POST 去续上任务，符合摘录规则。"
 },
 {
  "id": "T18",
  "answer_hash": "e52c3a124cd5",
  "correct": true,
  "reason": "答案明确说明样本必须在0–30秒内、不会裁剪超出部分，并且会拒绝作图，符合摘录规则且无矛盾。"
 },
 {
  "id": "T19",
  "answer_hash": "e0999f1a5af2",
  "correct": true,
  "reason": "答案明确指出不能用 --tail 100（--tail 为布尔开关）并给出了直接使用 --tail 实时跟随日志的正确用法。"
 },
 {
  "id": "T20",
  "answer_hash": "648b81bbbb62",
  "correct": true,
  "reason": "答案明确说明 start/clean 只有在机器人停靠底座时才有效，不在底座会回 NOT_READY（退出码 20），并给出先让其回座再开始清扫的做法，未与规则矛盾。"
 },
 {
  "id": "T21",
  "answer_hash": "d361b2704d5b",
  "correct": true,
  "reason": "答案明确说明需使用带末尾8位日期的 canonical_slug，并指出短 slug 会被 model-activity 拒绝并返回空数组。"
 },
 {
  "id": "T22",
  "answer_hash": "0c5bfe2a601b",
  "correct": true,
  "reason": "答案明确说明默认会包含子/后代会话，并给出只保留原始匹配需加 --no-expand，且无矛盾。"
 },
 {
  "id": "T23",
  "answer_hash": "727e1a36931f",
  "correct": true,
  "reason": "答案明确指出不要在测峰值前用 -ac 1 下混，解释了相关立体声约 +3 dB 导致误报削波并举出 -1.5 dBFS/1.19 的例子，同时给出了 ffmpeg -af ebur128=peak=true -f null - 用于交叉验证。"
 },
 {
  "id": "T24",
  "answer_hash": "daf32fee86ef",
  "correct": true,
  "reason": "明确给出普通句子建议为12–20词，并清楚说明硬性上限为22词，且未与规则相矛盾。"
 },
 {
  "id": "T25",
  "answer_hash": "6eacf8ab0fd0",
  "correct": true,
  "reason": "答案明确说明对 search/extract 使用 --stdout 会报未识别参数、退出码为 2，且不会发起网络请求，完全符合引用规则。"
 },
 {
  "id": "T26",
  "answer_hash": "734f5cda7e5a",
  "correct": true,
  "reason": "答案明确给出“每四小节最多一次”、必须登记在 skeleton_allow，且通常只在终止式的限制，并说明强拍（1与3拍）须与原谱音高类一致，符合摘录规则。"
 },
 {
  "id": "T27",
  "answer_hash": "1a8f93d84939",
  "correct": true,
  "reason": "答案明确说明了 --quality 在 Gemini 下被忽略且该参数只对 GPT Image 模型生效，完全符合摘录内容且无矛盾。"
 },
 {
  "id": "T28",
  "answer_hash": "a7fe1d089c7d",
  "correct": true,
  "reason": "答案明确给出：409 时不会创建；要并存需带 \"confirm_conflict\": true 重新提交；要替换需先用 POST /scheduled/{id}/cancel 取消旧任务再提交，新旧都与摘录一致且未矛盾。"
 },
 {
  "id": "T29",
  "answer_hash": "126090fea2c6",
  "correct": true,
  "reason": "答案明确指出 --after/--before 是 UUID 分页游标而非时间过滤器，不能传 ISO 时间，并给出了先导出最近邮件再按 created_at 本地过滤的正确流程。"
 },
 {
  "id": "T30",
  "answer_hash": "71b15d72dfde",
  "correct": true,
  "reason": "它明确说明文章分发默认用长帖且不强加280字限制，“一条帖子”指Typefully帖子，并指出要短推需加 --no-long-post。"
 }
]