test.jsonl
16.9 KB
{"input": "【场景：单平台发布】用户希望将一篇旅游攻略图文发布到小红书。原始内容为：标题'云南七日游终极攻略：从大理到西双版纳，看这一篇就够了'（28字），正文2500字，配图15张。请给出完整的发布执行方案。", "expected_output": {"platform": "xiaohongshu", "adapter": "rpa", "content_adaptation": {"title": "云南七日游攻略：从大理到西双版纳", "title_length": 17, "body": "800字，分6段，短句+Emoji，如☁️🌸🍜", "image_count": 9, "image_format": "1080x1440 (3:4)", "tags": ["#云南旅游", "#大理", "#西双版纳", "#旅游攻略", "#自由行"]}, "preprocess": {"dedup": "图片轻微裁剪+饱和度+5%", "aigc_label": "添加显著+隐式双重标识（若为AI辅助生成）"}, "execution": {"tool": "Playwright/Selenium模拟真人浏览器操作", "random_delay": "登录后30-120秒随机间隔再点击发布", "time_window": "20:00-22:00 小红书黄金时段"}, "state_tracking": {"task_id": "T-xhs-001", "status_flow": "PENDING->QUEUED->PUBLISHING->PUBLISHED", "retry": "RPA失败最多重试3次，指数退避", "failure_handling": "连续失败转FAILED_PERMANENT并告警"}}, "dimensions": {"platform_adaptation": 15, "content_preprocessing": 14, "publish_execution": 13, "state_tracking_retry": 13, "compliance_risk_control": 13, "overall_design": 20}, "score": 88, "grader_hints": "小红书单平台方案完整，适配准确，RPA行为模拟到位。但RPA未提IP隔离，风控扣2分；状态机缺少DLQ人工介入说明扣2分。"}
{"input": "【场景：多平台批量发布】某品牌有一组10条促销短视频，需要同时发布到抖音和视频号。请设计批量发布的系统执行流程。", "expected_output": {"platforms": ["douyin", "video_channel"], "preprocess": {"video": {"format": "竖屏9:16", "duration": "≤60分钟", "dedup": ["修改MD5", "调整帧率±1fps", "添加0.5秒随机转场", "加载3dB底噪"]}, "copy": "每条视频标题≤100字，描述轮换3套模板", "aigc_label": "抖音显著标识+视频号建议标识"}, "batch_config": {"chunk_size": 5, "interval": "600秒", "错峰": "抖音18:00-20:00，视频号20:00-22:00"}, "adapters": {"douyin": {"type": "api|rpa", "priority": "企业号API，个人号Playwright"}, "video_channel": {"type": "rpa", "tool": "social-auto-upload/tencent_uploader"}}, "queue": "分别进入douyin_queue和video_channel_queue，独立调度", "state_tracking": "每个视频x2平台=20个PublishTask，统一batch_id=BAT-2026-PROMO", "monitoring": "实时成功率看板，任平台失败独立重试不影响另一平台"}, "dimensions": {"platform_adaptation": 14, "content_preprocessing": 14, "publish_execution": 14, "state_tracking_retry": 14, "compliance_risk_control": 13, "overall_design": 22}, "score": 91, "grader_hints": "批量方案设计完善，去重、错峰、状态追踪都到位。视频号RPA稳定性未提降级策略扣1分；风控缺少IP隔离扣2分。"}
{"input": "【场景：定时任务】用户设置了一篇公众号文章在下周三早7:30定时发布。系统当前时间是周日。请描述从任务创建到最终发布的完整流程。", "expected_output": {"task_creation": {"scheduled_at": "下周三07:30", "platform": "wechat", "account_type": "订阅号"}, "preprocess": {"html": "完整保留，封面/thumb_media_id已预上传", "checklist": ["无JS", "<2万字", "图片<1MB且使用微信URL", "thumb_media_id有效"]}, "queue": "进入delayed_queue，延迟到周三07:30触发", "execution_at_publish_time": {"step1": "获取最新access_token（自动刷新）", "step2": "调用draft/add创建草稿", "step3": "调用freepublish/submit发布", "step4": "记录平台返回的publish_id"}, "state_tracking": {"status": "PENDING->QUEUED(delayed)->PUBLISHING->PUBLISHED", "fallback": "若发布失败且为可重试错误(500/503)，指数退避最多5次", "permanent_failure": "400/401立即停止并告警"}}, "dimensions": {"platform_adaptation": 14, "content_preprocessing": 12, "publish_execution": 14, "state_tracking_retry": 14, "compliance_risk_control": 11, "overall_design": 20}, "score": 85, "grader_hints": "定时任务流程清晰，延迟队列和token刷新都考虑到。但风控维度未涉及AIGC标识和预审（若内容为AI生成则有风险），扣4分。"}
{"input": "【场景：失败重试】系统向微博发布一条内容后收到HTTP 503错误，当前attempt_count=2，max_attempts=5。请给出接下来系统的处理逻辑。", "expected_output": {"error_classification": "可重试错误（服务端暂时不可用）", "status_transition": "PUBLISHING->FAILED_RETRYABLE", "retry_logic": {"backoff": "exponential + jitter", "calculation": "base=2^attempt * 1000ms ± 20% jitter", "next_retry_at": "attempt=2 -> 约4秒~6秒后重试", "max_attempts": 5}, "next_action": "将任务重新入队weibo_queue，等待调度器在next_retry_at后执行", "monitoring": "记录last_error_code=503，last_error_message=Service Unavailable", "escalation": "若attempt_count达到5仍未成功，转FAILED_PERMANENT并进入DLQ，通知运营"}, "dimensions": {"platform_adaptation": 10, "content_preprocessing": 8, "publish_execution": 13, "state_tracking_retry": 15, "compliance_risk_control": 9, "overall_design": 16}, "score": 71, "grader_hints": "状态机和重试策略完全正确，指数退避+jitter专业。但503是否可持续重试5次未考虑平台熔断策略，执行扣2分；综合未提熔断器，扣1分。"}
{"input": "【场景：内容预审】用户提交了一条带货文案，标题为'这款面膜真的绝绝子！姐妹们戳链接立减50元，库存只剩100件'，计划发布到小红书和微博。请给出预审结果和处置建议。", "expected_output": {"moderation_result": {"level": "B级高风险", "reason": "紧迫感营销（只剩100件）、未经验证的降价承诺、诱导点击"}, "fast_filter_hits": ["立减", "只剩", "戳链接"], "ai_review": "BERT判定为强营销内容，小红书极易限流，微博可能被投诉", "disposition": {"weibo": "降级修改后发布：删除紧迫感措辞，补充真实库存凭证", "xiaohongshu": "建议拦截或大幅修改：小红书严禁交易导流和强营销话术"}, "suggested_revision": {"title": "这款面膜使用分享，性价比确实不错", "body": "客观描述使用感受，不提具体库存和夸张折扣，链接改用平台认可的方式（如微博小店/小红书薯条）"}, "final_action": "AIGC标识（若内容由AI辅助生成）"}, "dimensions": {"platform_adaptation": 13, "content_preprocessing": 12, "publish_execution": 8, "state_tracking_retry": 8, "compliance_risk_control": 14, "overall_design": 14}, "score": 69, "grader_hints": "预审判断准确，平台差异处置合理。但状态追踪和执行维度因场景未涉及，只得基础分；小红书严禁交易导流，该直接拦截，风控扣1分。"}
{"input": "【场景：账号风控】系统在小红书RPA发布时，连续3条内容都因为'账号异常'而失败。请说明系统应该如何响应，以保护账号不被永久封禁。", "expected_output": {"immediate_action": "立即停止该账号所有queued和pending任务", "status_transition": "所有未执行任务转CANCELLED或FAILED_PERMANENT", "account_health_check": "标记该账号为SUSPICIOUS，禁止24小时内再次使用", "investigation_steps": ["1) 检查最近发布频率是否过高", "2) 检查内容重复率是否超标", "3) 检查IP/设备指纹是否异常", "4) 确认是否被平台风控系统命中"], "recovery_plan": {"short_term": "24小时后使用低风险内容测试1条", "long_term": "如测试成功，逐步恢复发布频率；如仍失败，弃用该账号", "prevention": ["增加随机延迟30-180秒", "每日发布量控制在平台安全阈值内", "IP轮换与设备隔离", "强化去重和AIGC标识"]}, "notification": "告警运营：账号xxx小红书异常，已自动停机保护"}, "dimensions": {"platform_adaptation": 12, "content_preprocessing": 12, "publish_execution": 12, "state_tracking_retry": 13, "compliance_risk_control": 14, "overall_design": 18}, "score": 81, "grader_hints": "账号保护响应及时，停机、调查、恢复、预防都有。但缺少DLQ和具体的人工复核流程，状态追踪扣2分；未给出SUSPICIOUS状态的数据模型设计，综合扣2分。"}
{"input": "【场景：新平台接入】公司希望在一周内将'Twitter/X'加入现有发布系统。现有系统使用Adapter+Pipeline架构。请给出最小可行接入方案。", "expected_output": {"approach": "新增TwitterAdapter实现统一Adapter接口", "adapter_methods": ["initialize(config)", "adaptContent(post)", "publish(task)", "parseResponse(rawResponse)", "refreshAuth()"], "twitter_specifics": {"auth": "OAuth2.0或OAuth1.0a", "limits": "推文≤280字符（会员4000），图片≤4张，视频≤2分20秒", "thread_support": "长推文拆分为thread链式发布"}, "implementation_steps": ["1) 实现twitter_adapter.js", "2) 在PlatformConfig中添加twitter规则", "3) 在AdapterRegistry注册", "4) 复用Pipeline/Queue/StateTracking核心代码"], "estimated_effort": "3-5个工作日（含测试）", "risks": "X平台API政策变动频繁，需监控官方公告"}, "dimensions": {"platform_adaptation": 13, "content_preprocessing": 10, "publish_execution": 12, "state_tracking_retry": 12, "compliance_risk_control": 10, "overall_design": 23}, "score": 80, "grader_hints": "扩展方案清晰，符合Adapter+配置化原则，3-5天估算合理。但Twitter具体的风控策略（如封号、限流阈值）未分析，平台适配和风控各扣1-2分。"}
{"input": "【场景：合规标识】系统检测到用户上传的视频是由AI生成的数字人讲解。该视频需要发布到抖音、微博、小红书。请说明各平台的AIGC标识要求和系统的注入策略。", "expected_output": {"platform_requirements": {"douyin": "强制显著标识：需在视频画面或描述中明确标注'本内容系AI生成'", "weibo": "建议声明：平台已增设'未添加AI标识'投诉分类，企业号应主动标注", "xiaohongshu": "强制双标识：画面显著标识+描述文字声明"}, "system_strategy": {"detection": "内置AIGC检测或读取用户声明is_aigc=true", "injection": {"video": "在视频开头/结尾叠加3秒'本内容系AI生成'水印", "description": "按平台规范追加AIGC声明文案", "metadata": "注入平台要求的隐式标签/元数据"}, "pre_publish_check": "每个平台发布前校验AIGC标识是否符合该平台2025年最新规范"}, "risk_if_missing": {"douyin": "限流或下架", "weibo": "被投诉后限流", "xiaohongshu": "封号风险"}}, "dimensions": {"platform_adaptation": 14, "content_preprocessing": 14, "publish_execution": 12, "state_tracking_retry": 9, "compliance_risk_control": 15, "overall_design": 18}, "score": 82, "grader_hints": "AIGC标识方案全面，覆盖了三个平台的差异。但状态追踪未涉及发布后对AIGC标识有效性的监控，扣3分；执行扣1分因为未提标识失败后的拦截策略。"}
{"input": "【场景：混合执行】一篇公司新闻稿需要发布到微博、微信公众号、抖音、小红书、视频号5个平台。请给出完整的端到端执行方案，包括Adapter选择、队列调度、状态追踪和失败处理。", "expected_output": {"adapters": {"weibo": "WeiboAPIAdapter (OAuth2.0)", "wechat": "WechatAPIAdapter (access_token)", "douyin": "DouyinAPIAdapter (企业) 或 DouyinRPAAdapter (个人)", "xiaohongshu": "XHSRPAAdapter (Playwright)", "video_channel": "VideoChannelRPAAdapter (tencent_uploader)"}, "pipeline": ["Ingest", "Preprocess (适配/去重/AIGC/预审)", "Moderation", "Queue (按平台拆分成5条子任务)", "Publish (错峰: 微博08:00, 微信12:00, 抖音18:00, 小红书20:00, 视频号22:00)", "Track"], "state_tracking": {"master_task": "MT-001", "sub_tasks": [{"task_id": "T-wb-001", "platform": "weibo"}, {"task_id": "T-wx-001", "platform": "wechat"}, {"task_id": "T-dy-001", "platform": "douyin"}, {"task_id": "T-xhs-001", "platform": "xiaohongshu"}, {"task_id": "T-vc-001", "platform": "video_channel"}], "aggregation": "MasterTask成功率=5/5时标记全部成功；任意失败不影响其他平台"}, "failure_handling": {"retryable": "429/500按平台指数退避重试", "permanent": "400/401/风控转FAILED_PERMANENT并进入DLQ", "notification": "实时推送各平台状态到运营看板"}}, "dimensions": {"platform_adaptation": 15, "content_preprocessing": 14, "publish_execution": 14, "state_tracking_retry": 14, "compliance_risk_control": 13, "overall_design": 23}, "score": 93, "grader_hints": "端到端方案非常完整，Adapter选择正确，错峰调度合理，状态聚合清晰。但RPA平台未提IP/设备隔离，风控扣2分；综合能力优秀但缺少配置热更新说明，扣1分。"}
{"input": "【场景：配置变更】微博在2025年Q2将个人用户日限从30条调整为20条，标题会员长度从5000字缩减为2000字。请说明系统如何在不停机的情况下响应这一规则变化。", "expected_output": {"response_flow": {"step1": "运营人员在管理后台修改PlatformConfig中weibo的daily_publish_limit=20和title_max_length=2000", "step2": "配置热更新服务将变更推送至所有发布节点", "step3": "Adapter在下次发布时自动读取新配置"}, "impact_mitigation": {"queued_tasks": "已入队任务按旧配置执行（避免中途变更导致不一致）", "new_tasks": "新提交任务立即按新配置校验和裁剪", "rate_limiter": "个人账号日限实时生效，第21条自动拒绝入队"}, "validation": "系统记录配置变更日志（config_version + timestamp），便于审计和回滚"}, "dimensions": {"platform_adaptation": 14, "content_preprocessing": 11, "publish_execution": 12, "state_tracking_retry": 10, "compliance_risk_control": 11, "overall_design": 23}, "score": 81, "grader_hints": "配置化响应方案合理，热更新和版本控制都有。但已入队任务是否应继续按旧配置执行存在争议（最好按新配置），平台适配扣1分；缺少对超长发内容的降级策略说明，综合扣2分。"}
{"input": "【场景：多媒体转码】用户上传了一个4K横屏视频（16:9，时长8分钟），需要发布到抖音和视频号。请描述系统的内容预处理过程。", "expected_output": {"input_specs": {"resolution": "3840x2160", "aspect_ratio": "16:9", "duration": "8分钟", "format": "mp4"}, "preprocess_steps": [{"step": "视频压缩与转码", "output": "1080p或720p，码率适配移动端"}, {"step": "画幅裁剪", "douyin": "9:16竖屏，中心裁剪+智能追踪", "video_channel": "9:16竖屏"}, {"step": "时长检查", "result": "8分钟 ≤ 抖音60分钟/视频号30分钟，无需裁剪"}, {"step": "封面提取", "action": "自动抽取第3秒作为封面"}, {"step": "MD5扰动", "action": "修改元数据改变MD5"}, {"step": "AIGC检查", "action": "若AI生成则注入标识"}], "quality_assurance": "转码后自动抽样播放校验5秒，确保音画同步"}, "dimensions": {"platform_adaptation": 14, "content_preprocessing": 15, "publish_execution": 11, "state_tracking_retry": 9, "compliance_risk_control": 11, "overall_design": 17}, "score": 77, "grader_hints": "多媒体预处理专业，覆盖转码、裁剪、封面、去重。但8分钟视频对于视频号已经接近上限，未提分段或压缩建议；状态追踪未涉及转码任务的状态管理，综合扣2分。"}
{"input": "【场景：夜间批量失败】凌晨2点，系统批量发布50条微博，其中30条成功，15条遇到429，5条遇到401。请给出第二天早上运营人员查看系统时，应该看到的状态报表和处理建议。", "expected_output": {"status_report": {"total": 50, "published": 30, "failed_retryable": 15, "failed_permanent": 5, "success_rate": "60%"}, "error_breakdown": {"429": "15条，原因：凌晨2点非黄金时段大量发布触发rate limit，当前在指数退避重试队列中", "401": "5条，原因：access_token于凌晨01:45过期，自动刷新机制异常"}, "action_items": [{"priority": "P0", "issue": "401批量失败", "action": "立即检查token刷新服务日志，修复后手动重试5条"}, {"priority": "P1", "issue": "429高峰", "action": "调整批量策略，将发布时间分布到08:00-10:00和20:00-22:00，避免集中涌入"}, {"priority": "P2", "issue": "成功率偏低", "action": "审查内容重复率，加强去重；检查账号健康度"}], "system_improvements": {"token_refresh": "增加过期前10分钟预刷新+多重告警", "rate_limit": "集成微博Rate-Limit头部读取，动态调整队列速度", "randomization": "批量任务增加30-300秒随机抖动"}}, "dimensions": {"platform_adaptation": 12, "content_preprocessing": 11, "publish_execution": 13, "state_tracking_retry": 14, "compliance_risk_control": 11, "overall_design": 19}, "score": 80, "grader_hints": "报表清晰，分类处理和改进建议都到位。但401失败5条应立即停止而不是继续发到5条失败，系统缺少熔断机制；综合扣2分。"}