Datasets:

Zhanli-Li
/

DataWorld-Train-2k-part2

name stringlengths 25 31	instruction stringlengths 232 1.35k	description stringlengths 30 66	keywords listlengths 4 4	schema_version stringclasses 1 value	config unknown	metadata unknown	files listlengths 17 9.17k	location stringlengths 25 31
dataworld/p000500-consequence-2	运营这边的菜单结构这两年调过两次：2023年10月加了一批低价套餐，去年12月又下架了其中几款。最近复盘我在纠结一件事——我们这几档套餐（低价的、常规的、高端的）到底在怎么影响生意？具体说，套餐的定价结构（组合价相对成本、相对单点价的折让幅度）跟单均实收是什么关系，跟客人打出的物超所值分又是什么关系。我尤其想知道那批低价套餐是不是在用客单价换满意度，这笔账到底划不划算。麻烦你把数据捋一遍，给我一个有依据的结论，别只看表面数字。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `env/xxx` 即 `/app/data/xxx`）。把你的完整回答写入 `/app/answer...	后果分析：菜单与套餐结构（data-world 世界 p000500）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000500-consequence-2", "description": "后果分析：菜单与套餐结构（data-world 世界 p000500）", ...	{ "difficulty_explanation": "第 2 层后果分析题：需在 p000500 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的后果推断结论。", "category": "data-analysis", "difficulty": "hard", "world": "p000500", "dag_nodes": 8, "dag_edges": 11, "dag_category_id": 2, "dag_category_name": "Customer, Pricing & Revenue / 客户·定价·收益" }	[ "environment/Dockerfile", "environment/data/crm/customer_feedback_events.csv", "environment/data/crm/customer_lifecycle_monthly.csv", "environment/data/crm/customer_master_scd.csv", "environment/data/finance/monthly_store_summary.xlsx", "environment/data/macro/regional_economic_indicators_monthly.csv", ...	dataworld-p000500-consequence-2
dataworld/p000502-cause-3	我们做儿童民谣流媒体快三年了，听众留存一直是最让我没底的事——首播一周后还在听的人少得可怜，一个月后更是所剩无几。下季度推广预算和产品迭代方向马上要定，我需要你把手头的数据捋一遍，把“到底是什么在驱动或预示听众的留存和召回”这件事给我讲明白：谁留得下来、谁会被召回回来、为什么，其中哪些是我们能信的真信号，哪些可能是数据本身把我们带偏了看错的。我不想看一堆相关系数罗列，要的是能拿来拍板的判断——哪个杠杆值得继续加码，哪个看着诱人其实不是那么回事。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `env/xxx` 即 `/app/data/xxx`）。把你的完整回答写入 `/a...	原因分析：留存与召回（data-world 世界 p000502）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000502-cause-3", "description": "原因分析：留存与召回（data-world 世界 p000502）", "author...	{ "difficulty_explanation": "第 3 层原因分析题：需在 p000502 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的因果归因结论。", "category": "data-analysis", "difficulty": "hard", "world": "p000502", "dag_nodes": 12, "dag_edges": 25, "dag_category_id": 1, "dag_category_name": "Commercial Growth & Acquisition / 商业增长与获客" }	[ "environment/Dockerfile", "environment/data/analytics/listener_daily_engagement.parquet", "environment/data/analytics/retention_weekly.csv", "environment/data/calendar/external_environment_daily.csv", "environment/data/catalog/releases.csv", "environment/data/catalog/tracks_scd.parquet", "environment/da...	dataworld-p000502-cause-3
dataworld/p000505-cause-1	我们这一年在外部 7 个获客渠道上砸了不少预算，装量是有了，但我现在分不清哪些钱花得值、哪些其实在打水漂。你帮我把渠道触点这层数据捋一捋：到底是投得越多装得越多，还是素材本身的好坏也在起作用？哪些渠道的安装效率真的更高？有没有什么指标看着漂亮、实际却在坑我们？另外数据里要是有会把我带偏的坑，也一并点出来——我接下来要拿你的结论去调预算，不想踩雷。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `env/xxx` 即 `/app/data/xxx`）。把你的完整回答写入 `/app/answer.txt`。	原因分析：渠道触点流量（data-world 世界 p000505）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000505-cause-1", "description": "原因分析：渠道触点流量（data-world 世界 p000505）", "autho...	{ "difficulty_explanation": "第 1 层原因分析题：需在 p000505 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的因果归因结论。", "category": "data-analysis", "difficulty": "medium", "world": "p000505", "dag_nodes": 12, "dag_edges": 20, "dag_category_id": 1, "dag_category_name": "Commercial Growth & Acquisition / 商业增长与获客" }	[ "environment/Dockerfile", "environment/data/_meta/data_dictionary.md", "environment/data/_meta/export_manifest.json", "environment/data/ad_ops/creative_library.csv", "environment/data/ad_ops/daily_channel_spend.csv", "environment/data/attribution/daily_touchpoint_traffic.tsv", "environment/data/billing/...	dataworld-p000505-cause-1
dataworld/p000505-cause-2	今年新客数据有点看不懂，想让你帮我理一理。我们这一年改过一次注册流程（3 月那次），新客注册量和激活量前后起伏很大；投放这边七个渠道、十几套定向人群码一直在跑，老板让我判断到底哪些动作真正把新客"拉进来并激活"了，哪些渠道值得加预算。具体我想知道两件事：一是装了我们 App 的人最终有没有注册、注册了的人有没有真正激活（首次进游戏或看剧），这中间是什么在卡人、什么在推动人；二是各获客渠道和定向人群在激活上到底差在哪，给我一个能落地的渠道/定向建议。过程中要是发现某些看起来很顺的结论其实站不住脚，也一并点出来，别让我拿着错判断去调预算。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面...	原因分析：新客注册激活（data-world 世界 p000505）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000505-cause-2", "description": "原因分析：新客注册激活（data-world 世界 p000505）", "autho...	{ "difficulty_explanation": "第 2 层原因分析题：需在 p000505 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的因果归因结论。", "category": "data-analysis", "difficulty": "hard", "world": "p000505", "dag_nodes": 12, "dag_edges": 20, "dag_category_id": 1, "dag_category_name": "Commercial Growth & Acquisition / 商业增长与获客" }	[ "environment/Dockerfile", "environment/data/_meta/data_dictionary.md", "environment/data/_meta/export_manifest.json", "environment/data/ad_ops/creative_library.csv", "environment/data/ad_ops/daily_channel_spend.csv", "environment/data/attribution/daily_touchpoint_traffic.tsv", "environment/data/billing/...	dataworld-p000505-cause-2
dataworld/p000505-consequence-2	我们最近在游戏侧做了不少促活的动作，游戏参与整体往上走，老板让我搞清楚一件事：用户在游戏里玩得越深，后面到底会发生什么。具体说，玩得深的用户会不会更频繁地切去看剧休息？他们的社交互动（观影派对、成就分享、社区这些）是更活跃，还是反而变少？最关键的是，玩得深到底有没有转化成花钱，尤其是游戏内购——如果有，是普遍在花还是集中在少数人身上？我想看到方向和大致的量级，以及哪些用户群体上表现不一样。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `env/xxx` 即 `/app/data/xxx`）。把你的完整回答写入 `/app/answer.txt`。	后果分析：游戏参与深度（data-world 世界 p000505）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000505-consequence-2", "description": "后果分析：游戏参与深度（data-world 世界 p000505）", ...	{ "difficulty_explanation": "第 2 层后果分析题：需在 p000505 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的后果推断结论。", "category": "data-analysis", "difficulty": "hard", "world": "p000505", "dag_nodes": 12, "dag_edges": 20, "dag_category_id": 1, "dag_category_name": "Commercial Growth & Acquisition / 商业增长与获客" }	[ "environment/Dockerfile", "environment/data/_meta/data_dictionary.md", "environment/data/_meta/export_manifest.json", "environment/data/ad_ops/creative_library.csv", "environment/data/ad_ops/daily_channel_spend.csv", "environment/data/attribution/daily_touchpoint_traffic.tsv", "environment/data/billing/...	dataworld-p000505-consequence-2
dataworld/p000506-cause-2	我们这个文学遗产纵向面板马上要结题，数据室那边催着要一份关于「启蒙价值观量表」的归因小结。六轮问卷（Q10 理性主义、Q11 言论自由、Q12 个人主义三个分项）从 2021 跑到 2023，我想搞清楚到底是谁在抬这些分、谁在压它们——是人口背景（教育、城乡、收入）起决定作用，还是对 18 世纪经典文学的接触在起作用？文学接触这条线到底是越多越高，还是有拐点？顺便帮我把「看着像信号、其实是坑」的地方挑出来：哪些波动是流失造成的、哪些是问卷模式或量表换版造成的，别让结论被这些带偏。我只要结论和它站得住的证据，不需要面面俱到。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `en...	原因分析：启蒙价值观量表得分（data-world 世界 p000506）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000506-cause-2", "description": "原因分析：启蒙价值观量表得分（data-world 世界 p000506）", "au...	{ "difficulty_explanation": "第 2 层原因分析题：需在 p000506 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的因果归因结论。", "category": "data-analysis", "difficulty": "hard", "world": "p000506", "dag_nodes": 9, "dag_edges": 13, "dag_category_id": 15, "dag_category_name": "Social Science, Survey & Behavioral / 社会科学·调查·...	[ "environment/Dockerfile", "environment/data/algorithm/daily_push/daily_push_2021.parquet", "environment/data/algorithm/daily_push/daily_push_2022.parquet", "environment/data/algorithm/daily_push/daily_push_2023.parquet", "environment/data/algorithm/daily_push/daily_push_2024H1.parquet", "environment/data/...	dataworld-p000506-cause-2
dataworld/p000507-cause-2	生产这边的数我看了，总觉得不对劲。今年（2024）的产线产出明显不如去年，厂长会上让我拿数据说清楚到底差在哪、是什么把实际产出压下去了，别只给我一句"设备老了"。MES 里的批次产出、设备状态、还有排程计划都在数据里，你帮我捋一捋：这两年实际产出相对计划到底掉了多少，主要被哪些因素吃掉了，这些因素之间谁更主导、谁只是看着像。重点放在能从批次和设备记录里坐实的证据上，别拿没数支撑的猜测凑结论。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `env/xxx` 即 `/app/data/xxx`）。把你的完整回答写入 `/app/answer.txt`。	原因分析：生产执行与产出（data-world 世界 p000507）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000507-cause-2", "description": "原因分析：生产执行与产出（data-world 世界 p000507）", "auth...	{ "difficulty_explanation": "第 2 层原因分析题：需在 p000507 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的因果归因结论。", "category": "data-analysis", "difficulty": "hard", "world": "p000507", "dag_nodes": 11, "dag_edges": 15, "dag_category_id": 4, "dag_category_name": "Operations, Supply Chain & Manufacturing / 运营·...	[ "environment/Dockerfile", "environment/data/crm/customer_satisfaction_weekly.csv", "environment/data/crm/orders_raw.jsonl", "environment/data/erp/purchase_orders.csv", "environment/data/master/bom_master.csv", "environment/data/master/carrier_master.csv", "environment/data/master/customer_master.csv", ...	dataworld-p000507-cause-2
dataworld/p000507-cause-4	这两年我们的发运在途时效越来越难看——日常两三天，到 2024 下半年和每次大促前后动不动四五天往上，客户已经在催了。物流那边一直说是承运商不行、运力不够，想让我批钱换承运商、加运力。我不太信这个说法，但也没数据反驳。你把发运那套数据从头到尾捋一遍，告诉我：到底什么在拖慢我们的在途时效？是承运商的问题，还是别的？哪些因素是真能解释时效变化的、哪些只是看着像。最后给我一个能站住脚的判断：如果真要改善时效，该往哪个方向使劲。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `env/xxx` 即 `/app/data/xxx`）。把你的完整回答写入 `/app/answer.tx...	原因分析：物流发运与时效（data-world 世界 p000507）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000507-cause-4", "description": "原因分析：物流发运与时效（data-world 世界 p000507）", "auth...	{ "difficulty_explanation": "第 4 层原因分析题：需在 p000507 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的因果归因结论。", "category": "data-analysis", "difficulty": "hard", "world": "p000507", "dag_nodes": 11, "dag_edges": 15, "dag_category_id": 4, "dag_category_name": "Operations, Supply Chain & Manufacturing / 运营·...	[ "environment/Dockerfile", "environment/data/crm/customer_satisfaction_weekly.csv", "environment/data/crm/orders_raw.jsonl", "environment/data/erp/purchase_orders.csv", "environment/data/master/bom_master.csv", "environment/data/master/carrier_master.csv", "environment/data/master/customer_master.csv", ...	dataworld-p000507-cause-4
dataworld/p000507-consequence-2	我每周锁 MPS、给产线下发日批次排程，可车间说排的量做不出来、销售说该到的货不到，两头都骂排程。我想弄明白：我们排下去的生产计划到底有多少真变成了产出，缺口主要卡在哪，排程本身有没有在添乱，这两年是不是越来越差。你拿排程和实际执行对一下，顺着往后看到库存和客户履约，给我一个有数有据的判断。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `env/xxx` 即 `/app/data/xxx`）。把你的完整回答写入 `/app/answer.txt`。	后果分析：生产排程（data-world 世界 p000507）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000507-consequence-2", "description": "后果分析：生产排程（data-world 世界 p000507）", "a...	{ "difficulty_explanation": "第 2 层后果分析题：需在 p000507 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的后果推断结论。", "category": "data-analysis", "difficulty": "hard", "world": "p000507", "dag_nodes": 11, "dag_edges": 15, "dag_category_id": 4, "dag_category_name": "Operations, Supply Chain & Manufacturing / 运营·...	[ "environment/Dockerfile", "environment/data/crm/customer_satisfaction_weekly.csv", "environment/data/crm/orders_raw.jsonl", "environment/data/erp/purchase_orders.csv", "environment/data/master/bom_master.csv", "environment/data/master/carrier_master.csv", "environment/data/master/customer_master.csv", ...	dataworld-p000507-consequence-2
dataworld/p000512-consequence-1	市规划办那边让我牵头理一件事：我们这127个街区，十几年下来，社会经济底子不同的街区，后来的际遇差了多少？我想搞清楚一个街区的贫困程度跟它后来的基础设施状况、能拿到的城市更新投入、还有警力部署到底什么关系——方向上是不是越穷越糟，量级多大，有没有哪个区间例外或者反过来。下个月规划委员会要听一个说法，你帮我把这条线从数据里摸出来，给我一个能站得住的结论，哪些有把握、哪些因为数据本身的毛病说不死，也一并讲清楚。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `env/xxx` 即 `/app/data/xxx`）。把你的完整回答写入 `/app/answer.txt`。	后果分析：街区社会经济基线（data-world 世界 p000512）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000512-consequence-1", "description": "后果分析：街区社会经济基线（data-world 世界 p000512）", ...	{ "difficulty_explanation": "第 1 层后果分析题：需在 p000512 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的后果推断结论。", "category": "data-analysis", "difficulty": "medium", "world": "p000512", "dag_nodes": 11, "dag_edges": 18, "dag_category_id": 13, "dag_category_name": "Public Policy & Program Evaluation / 公共政策与项...	[ "environment/Dockerfile", "environment/data/census/acs_methodology_notes.txt", "environment/data/census/migration_flows_annual.csv", "environment/data/census/tract_boundary_crosswalk_2010.csv", "environment/data/census/tract_socioeconomic_baseline_scd.csv", "environment/data/gis/infrastructure_survey_annu...	dataworld-p000512-consequence-1
dataworld/p000514-cause-1	我们这季想做更精准的拉新，市场那边吵了半天也没结论：到底什么样的人在我们合作流媒体上看过的 B 级片最多？是年轻的、有钱有闲的，还是别的什么人群？我手头有面板成员的人口档案，也有流媒体那边的观看汇总，但我自己不熟这批数据。帮我查清楚：哪些面板特征和一个人看过的 B 级片数量挂钩、影响各有多大，哪些看着相关其实经不起推敲。下周我要拿结论去定拉新目标人群，所以别只甩一堆相关系数给我——说清楚谁最值得拉、凭什么，以及你哪里没把握。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `env/xxx` 即 `/app/data/xxx`）。把你的完整回答写入 `/app/answe...	原因分析：邪典媒体曝光量（data-world 世界 p000514）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000514-cause-1", "description": "原因分析：邪典媒体曝光量（data-world 世界 p000514）", "auth...	{ "difficulty_explanation": "第 1 层原因分析题：需在 p000514 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的因果归因结论。", "category": "data-analysis", "difficulty": "medium", "world": "p000514", "dag_nodes": 8, "dag_edges": 10, "dag_category_id": 15, "dag_category_name": "Social Science, Survey & Behavioral / 社会科学·调...	[ "environment/Dockerfile", "environment/data/docs/panelist_welcome_guide.txt", "environment/data/docs/survey_protocol.md", "environment/data/panel_crm/panelist_profiles.csv", "environment/data/panel_crm/recruitment_log.csv", "environment/data/panel_crm/retention_followup.csv", "environment/data/social/mo...	dataworld-p000514-cause-1
dataworld/p000514-consequence-2	我下周要给产品组复盘这六轮调查的体验数据，想先弄清楚一件事：受访者对那批B级片的初始打分，到底会不会影响他们在后续任务环节里的行为。一个人给片子打完分之后，他对这些片子的态度能不能预测他接下来在这道题上花多长时间、会不会做到一半放弃、以及最后排顺序时来回改多少次？我不太确定这些关系是不是在六轮里都成立，听说后面几轮题目变长、有人做到一半就崩了，所以也想让你看看是不是有些轮次里头这套规律就看不出来了。方向、强弱、在哪些地方成立或失效，给我一个能站得住的判断就行，别只甩一个相关系数给我。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `env/xxx` 即 `/app/data/...	后果分析：初始态度与评价（data-world 世界 p000514）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000514-consequence-2", "description": "后果分析：初始态度与评价（data-world 世界 p000514）", ...	{ "difficulty_explanation": "第 2 层后果分析题：需在 p000514 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的后果推断结论。", "category": "data-analysis", "difficulty": "hard", "world": "p000514", "dag_nodes": 8, "dag_edges": 10, "dag_category_id": 15, "dag_category_name": "Social Science, Survey & Behavioral / 社会科学·调查·...	[ "environment/Dockerfile", "environment/data/docs/panelist_welcome_guide.txt", "environment/data/docs/survey_protocol.md", "environment/data/panel_crm/panelist_profiles.csv", "environment/data/panel_crm/recruitment_log.csv", "environment/data/panel_crm/retention_followup.csv", "environment/data/social/mo...	dataworld-p000514-consequence-2
dataworld/p000515-cause-2	我是市长刑事司法服务办公室的分析主管。Millbridge 2008 到 2019 这十二年，严重犯罪报案量涨了四成多，市议会下个月要我们交一份报告，说清楚到底是什么在推高报案、又被哪些因素牵动。手头有警情、311 市政工单、警务月报、社区信任调查和人口普查几套数据，分属不同部门，口径和格式都不统一。麻烦你把这些数据理顺，判断严重犯罪报案率到底被哪些因素驱动或预示——哪些是实打实的证据、哪些只是看着相关其实另有原因，哪些地方数据本身有坑、结论得打折扣。结论要拿去议会答辩，每个判断后面都得有能复算的数撑着。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `env/xxx` 即 ...	原因分析：严重犯罪报案率（data-world 世界 p000515）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000515-cause-2", "description": "原因分析：严重犯罪报案率（data-world 世界 p000515）", "auth...	{ "difficulty_explanation": "第 2 层原因分析题：需在 p000515 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的因果归因结论。", "category": "data-analysis", "difficulty": "hard", "world": "p000515", "dag_nodes": 9, "dag_edges": 13, "dag_category_id": 13, "dag_category_name": "Public Policy & Program Evaluation / 公共政策与项目评估...	[ "environment/Dockerfile", "environment/data/_raw/cad_daily_extract_sample.log", "environment/data/_raw/mocjs_program_config.kv", "environment/data/census/acs_tract_estimates_2008_2019.csv", "environment/data/census/migration_estimates_2008_2019.csv", "environment/data/corrections/offender_release_records_...	dataworld-p000515-cause-2
dataworld/p000515-cause-4	Millbridge 这十二年一直在掉人——全市从 2008 年的 15.26 万掉到 2019 年的 14.06 万。规划办想搞清楚居民到底为什么走、哪些街区在真正失血，好决定下一轮治安和住房干预往哪投。我手头有一份按街区年度统计的迁出/迁入估计，还有犯罪报案、矫正释放这些旁证数据。你别只给我一个总趋势——我要知道迁出背后真正在起作用的是什么，哪些证据靠得住、哪些看着像但其实不是。给我一个能站得住的判断。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `env/xxx` 即 `/app/data/xxx`）。把你的完整回答写入 `/app/answer.txt`。	原因分析：居民搬迁与人口流动（data-world 世界 p000515）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000515-cause-4", "description": "原因分析：居民搬迁与人口流动（data-world 世界 p000515）", "au...	{ "difficulty_explanation": "第 4 层原因分析题：需在 p000515 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的因果归因结论。", "category": "data-analysis", "difficulty": "hard", "world": "p000515", "dag_nodes": 9, "dag_edges": 13, "dag_category_id": 13, "dag_category_name": "Public Policy & Program Evaluation / 公共政策与项目评估...	[ "environment/Dockerfile", "environment/data/_raw/cad_daily_extract_sample.log", "environment/data/_raw/mocjs_program_config.kv", "environment/data/census/acs_tract_estimates_2008_2019.csv", "environment/data/census/migration_estimates_2008_2019.csv", "environment/data/corrections/offender_release_records_...	dataworld-p000515-cause-4
dataworld/p000516-consequence-4	这批面板跨了 2020、2022、2024 三次大选，我想要一个判断：第三方选民的票，到底换来了多少代表性产出——能不能换到下一周期的自动 ballot access、辩论资格、候选人上不上得了 ballot。这条转换在哪些州走得通、哪些州卡住，三年是走强还是走弱，背后主要是阈值门槛在起作用还是别的因素。给我方向、量级，还有能撑住结论的数据，别只甩个结论给我。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `env/xxx` 即 `/app/data/xxx`）。把你的完整回答写入 `/app/answer.txt`。	后果分析：投票行为与选举参与（data-world 世界 p000516）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000516-consequence-4", "description": "后果分析：投票行为与选举参与（data-world 世界 p000516）", ...	{ "difficulty_explanation": "第 4 层后果分析题：需在 p000516 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的后果推断结论。", "category": "data-analysis", "difficulty": "hard", "world": "p000516", "dag_nodes": 12, "dag_edges": 17, "dag_category_id": 15, "dag_category_name": "Social Science, Survey & Behavioral / 社会科学·调查...	[ "environment/Dockerfile", "environment/data/campaign/campaign_committees.parquet", "environment/data/campaign/digital_ad_exposure_log.parquet", "environment/data/campaign/precinct_competitiveness.parquet", "environment/data/campaign/targeting_pool_roster.parquet", "environment/data/campaign/voter_contact_...	dataworld-p000516-consequence-4
dataworld/p000518-cause-1	DEV-001 那台 UV-Vis 今年标定次数比去年翻了一番，是五台仪器里变化最扎眼的。我下周一要跟实验室主任交代这台仪器标定负担加重的原因——到底是仪器自身状态变了、排程在合理响应，还是排程本身出了问题。顺手帮我看看全实验室的标定质量：这两年下来的标定残差能不能接受，有没有哪台仪器或者我们用的参考物质在系统性地卡住标定能做到多准。标定记录、设备老化状态、参考物质目录、维护日志都在数据目录里，你自己拼。我要的是结论加依据，别把整理过程当结论端上来。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `env/xxx` 即 `/app/data/xxx`）。把你的完整回答写...	原因分析：仪器标定状态（data-world 世界 p000518）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000518-cause-1", "description": "原因分析：仪器标定状态（data-world 世界 p000518）", "autho...	{ "difficulty_explanation": "第 1 层原因分析题：需在 p000518 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的因果归因结论。", "category": "data-analysis", "difficulty": "medium", "world": "p000518", "dag_nodes": 12, "dag_edges": 15, "dag_category_id": 10, "dag_category_name": "Experimental & Physical Science / 实验与物理科学" ...	[ "environment/Dockerfile", "environment/data/analysis_platform/final_analysis_results.csv", "environment/data/analysis_platform/outlier_rejection_results.parquet", "environment/data/analysis_platform/replicate_variance.csv", "environment/data/analysis_platform/run_processing_map.csv", "environment/data/ana...	dataworld-p000518-cause-1
dataworld/p000518-consequence-1	最近几次质控复盘，好几起批次失控都扎堆在夏季高湿那几个月，技术主管怀疑是仪器室的环境（温度、湿度）在作怪。他想让你从这两年的运行数据里把这件事说清楚：仪器室的环境条件到底在多大程度上影响了测量读数和读数质量，方向是怎样的、量级有多大。实验室有五台光谱仪，原理各不相同，他特意交代别笼统下结论——哪台对环境最敏感、哪台最不敏感，要分开说清楚。制样室那边的环境会不会也通过样品制备环节影响到最终出结果，他没什么把握，让你顺手也看一眼，给个判断就行。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `env/xxx` 即 `/app/data/xxx`）。把你的完整回答写入 `/a...	后果分析：环境条件（data-world 世界 p000518）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000518-consequence-1", "description": "后果分析：环境条件（data-world 世界 p000518）", "a...	{ "difficulty_explanation": "第 1 层后果分析题：需在 p000518 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的后果推断结论。", "category": "data-analysis", "difficulty": "medium", "world": "p000518", "dag_nodes": 12, "dag_edges": 15, "dag_category_id": 10, "dag_category_name": "Experimental & Physical Science / 实验与物理科学" ...	[ "environment/Dockerfile", "environment/data/analysis_platform/final_analysis_results.csv", "environment/data/analysis_platform/outlier_rejection_results.parquet", "environment/data/analysis_platform/replicate_variance.csv", "environment/data/analysis_platform/run_processing_map.csv", "environment/data/ana...	dataworld-p000518-consequence-1
dataworld/p000518-consequence-3	小张，下季度我们要把异常读数剔除这一步写进测量不确定度评估的程序文件里，但在定稿前我需要你先把现状摸清楚。这两年我们出的每一份报告值和扩展不确定度，都是在剔除离群读数之后才算出来的，我一直担心两件事：一是剔除这步到底把我们报给客户的数值往哪个方向、改了多少；二是那些剔除比例特别高的批次，剔除之后不确定度是不是反而被压得太好看、没反映真实分散度。你从原始读数和剔除结果一路追到重复性方差和最终报告，把剔除这一步对报告值和不确定度的影响量化出来，高剔除批次单独看。哪些设备或情形下剔除更激进、更不可靠，也一并告诉我。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `env/xxx` ...	后果分析：异常剔除结果（data-world 世界 p000518）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000518-consequence-3", "description": "后果分析：异常剔除结果（data-world 世界 p000518）", ...	{ "difficulty_explanation": "第 3 层后果分析题：需在 p000518 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的后果推断结论。", "category": "data-analysis", "difficulty": "hard", "world": "p000518", "dag_nodes": 12, "dag_edges": 15, "dag_category_id": 10, "dag_category_name": "Experimental & Physical Science / 实验与物理科学" }	[ "environment/Dockerfile", "environment/data/analysis_platform/final_analysis_results.csv", "environment/data/analysis_platform/outlier_rejection_results.parquet", "environment/data/analysis_platform/replicate_variance.csv", "environment/data/analysis_platform/run_processing_map.csv", "environment/data/ana...	dataworld-p000518-consequence-3
dataworld/p000520-cause-4	过去两年我们签了四十多个艺人，到现在已经走了10个，还有几个挂在 at_risk 状态上。我需要搞清楚到底是什么在驱动艺人流失——是播放量不行、收入太低，还是别的。版税结算和播放数据都在数据目录里，艺人当前和历史状态也有记录。我不想看一份把所有变量都 corr 一遍的泛泛报告。给我能站住脚的结论：哪些因素真的预示或导致流失、量级多大，哪些看着相关其实是被别的东西带偏的或者口径算错了。如果有艺人明明数据很差却没走、或者数据还行却走了，把这种反例也讲清楚。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `env/xxx` 即 `/app/data/xxx`）。把你的完整回答...	原因分析：Artist Retention & Reactivation（data-world 世界 p000520）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000520-cause-4", "description": "原因分析：Artist Retention & Reactivation（data-world...	{ "difficulty_explanation": "第 4 层原因分析题：需在 p000520 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的因果归因结论。", "category": "data-analysis", "difficulty": "hard", "world": "p000520", "dag_nodes": 11, "dag_edges": 13, "dag_category_id": 1, "dag_category_name": "Commercial Growth & Acquisition / 商业增长与获客" }	[ "environment/Dockerfile", "environment/data/analytics/daily_streams.parquet", "environment/data/analytics/market_trends_weekly.csv", "environment/data/crm/artist_profiles.csv", "environment/data/crm/artist_status_history.csv", "environment/data/crm/outreach_events.csv", "environment/data/crm/production_...	dataworld-p000520-cause-4
dataworld/p000520-consequence-1	A&R 的 Lin 发来一封邮件：最近两个季度好几个流派在后台趋势看板上窜得很快，但我拿不准我们到底有没有跟上——scouting 是不是真的往热起来的流派偏了，marketing 预算又有没有跟着走。我想搞清楚：当一个流派在社交媒体和流媒体上的热度动起来之后，我们这边的反应是怎样的——scouting 外联和推广投放分别多久才跟上来、力度多大、有没有被什么稀释掉。给我一份能说清方向、快慢和量级的结论，我好判断接下来要不要调整节奏。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `env/xxx` 即 `/app/data/xxx`）。把你的完整回答写入 `/app/a...	后果分析：Market & Genre Trends（data-world 世界 p000520）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000520-consequence-1", "description": "后果分析：Market & Genre Trends（data-world 世界 ...	{ "difficulty_explanation": "第 1 层后果分析题：需在 p000520 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的后果推断结论。", "category": "data-analysis", "difficulty": "medium", "world": "p000520", "dag_nodes": 11, "dag_edges": 13, "dag_category_id": 1, "dag_category_name": "Commercial Growth & Acquisition / 商业增长与获客" }	[ "environment/Dockerfile", "environment/data/analytics/daily_streams.parquet", "environment/data/analytics/market_trends_weekly.csv", "environment/data/crm/artist_profiles.csv", "environment/data/crm/artist_status_history.csv", "environment/data/crm/outreach_events.csv", "environment/data/crm/production_...	dataworld-p000520-consequence-1
dataworld/p000523-cause-1	帮我看一下广告投放这块。这两年我们的总投放涨了一倍多，但每个渠道、每个地区的单次展现成本（CPM）和点击率（CTR）月度之间波动很大，有的渠道涨有的渠道反而降，财务那边一直追问我钱到底花在哪了、是不是买贵了。我想搞清楚一件事：到底是什么在驱动我们每个展现的成本和点击率——是我们自己买的量（投放量级）在变，还是市场上有什么外部因素在推？哪些是我们能控的，哪些是市场强加给我们的？渠道之间和地区之间是不是一样的逻辑，还是各走各的？给我一个有依据的判断，别只看个总量趋势就下结论。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `env/xxx` 即 `/app/data/xxx`）。...	原因分析：获客投放与触达（data-world 世界 p000523）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000523-cause-1", "description": "原因分析：获客投放与触达（data-world 世界 p000523）", "auth...	{ "difficulty_explanation": "第 1 层原因分析题：需在 p000523 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的因果归因结论。", "category": "data-analysis", "difficulty": "medium", "world": "p000523", "dag_nodes": 9, "dag_edges": 12, "dag_category_id": 1, "dag_category_name": "Commercial Growth & Acquisition / 商业增长与获客" }	[ "environment/Dockerfile", "environment/data/ad_platform/creative_registry.csv", "environment/data/ad_platform/daily_ad_performance.csv", "environment/data/ad_platform/monthly_market_index.csv", "environment/data/analytics/daily_user_activity.parquet", "environment/data/analytics/landing_sessions.jsonl", ...	dataworld-p000523-cause-1
dataworld/p000523-cause-2	最近两个季度新用户激活一直在好转——time-to-first-value 明显缩短，核心动作完成度也在抬。老板让我搞清楚这背后到底是什么在起作用：是我们这一年上的几个 onboarding 实验真有效，还是落地页路由调整带来的，又或者只是进来的用户结构变了。哪些是能复算的实打实的驱动、哪些是看着像但其实站不住，我得有个数，下半年才知道该往哪投。用这 21 个月的数据给我个结论。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `env/xxx` 即 `/app/data/xxx`）。把你的完整回答写入 `/app/answer.txt`。	原因分析：新用户激活（data-world 世界 p000523）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000523-cause-2", "description": "原因分析：新用户激活（data-world 世界 p000523）", "author...	{ "difficulty_explanation": "第 2 层原因分析题：需在 p000523 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的因果归因结论。", "category": "data-analysis", "difficulty": "hard", "world": "p000523", "dag_nodes": 9, "dag_edges": 12, "dag_category_id": 1, "dag_category_name": "Commercial Growth & Acquisition / 商业增长与获客" }	[ "environment/Dockerfile", "environment/data/ad_platform/creative_registry.csv", "environment/data/ad_platform/daily_ad_performance.csv", "environment/data/ad_platform/monthly_market_index.csv", "environment/data/analytics/daily_user_activity.parquet", "environment/data/analytics/landing_sessions.jsonl", ...	dataworld-p000523-cause-2
dataworld/p000523-cause-3	我们是做游戏分析 SaaS 的 PixelArc。这两年注册用户涨了不少，但董事会现在盯的是留存——新用户到底留不留得住、哪些东西能预示他们留得下来。你手上有一份每个用户每天的活跃面板，其余用户维度的记录也都在数据目录里。帮我把"是什么决定一个新用户留不留下来"讲清楚：哪些因素和次日及早期留存强相关、量级有多大；最近这几个季度上的功能，到底有没有把留存带起来。给我一份有依据的结论，数据里有什么坑你自己掂量。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `env/xxx` 即 `/app/data/xxx`）。把你的完整回答写入 `/app/answer.txt`。	原因分析：留存与活跃（data-world 世界 p000523）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000523-cause-3", "description": "原因分析：留存与活跃（data-world 世界 p000523）", "author...	{ "difficulty_explanation": "第 3 层原因分析题：需在 p000523 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的因果归因结论。", "category": "data-analysis", "difficulty": "hard", "world": "p000523", "dag_nodes": 9, "dag_edges": 12, "dag_category_id": 1, "dag_category_name": "Commercial Growth & Acquisition / 商业增长与获客" }	[ "environment/Dockerfile", "environment/data/ad_platform/creative_registry.csv", "environment/data/ad_platform/daily_ad_performance.csv", "environment/data/ad_platform/monthly_market_index.csv", "environment/data/analytics/daily_user_activity.parquet", "environment/data/analytics/landing_sessions.jsonl", ...	dataworld-p000523-cause-3
dataworld/p000524-cause-4	老规矩，麻烦你帮我看一下我们这 6 个组合的风险数据。风控复盘 2022 年时发现一件怪事：那年组合的 CVaR（条件回撤，尾部那块）涨得跟 VaR 不成比例，而且几个组合像约好了一起动。我想要个说法：到底是什么在驱动这 6 个组合的风险敞口（VaR 和 CVaR）？是单纯市场波动率上来了，还是另有原因？这风险在 6 个组合之间是均匀的，还是集中在某几个上？数据在投资组合那套文件里，相关的行情你自己找一下对上。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `env/xxx` 即 `/app/data/xxx`）。把你的完整回答写入 `/app/answer.txt`。	原因分析：投资组合与风险敞口（data-world 世界 p000524）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000524-cause-4", "description": "原因分析：投资组合与风险敞口（data-world 世界 p000524）", "au...	{ "difficulty_explanation": "第 4 层原因分析题：需在 p000524 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的因果归因结论。", "category": "data-analysis", "difficulty": "hard", "world": "p000524", "dag_nodes": 9, "dag_edges": 12, "dag_category_id": 3, "dag_category_name": "Finance, FP&A & Markets / 财务·FP&A·金融市场" }	[ "environment/Dockerfile", "environment/data/crm/orders/orders_2016.csv", "environment/data/crm/orders/orders_2017.csv", "environment/data/crm/orders/orders_2018.csv", "environment/data/crm/orders/orders_2019.csv", "environment/data/crm/orders/orders_2020.csv", "environment/data/crm/orders/orders_2021.cs...	dataworld-p000524-cause-4
dataworld/p000524-consequence-3	华锐铜业这八年铜价上蹿下跳，capex 却一直没停。我下周一要跟董事会汇报这家公司的财务健康度，需要你先把账理清楚：这八年下来，公司的杠杆、偿债能力和经营现金流到底是变好了还是变差了，主要被什么因素推着走？另外我尤其想搞明白一件事——这些财务上的好坏，到底有没有真正反映到股价上，还是说股价基本就跟着铜价走、资产负债表怎么样市场其实并不在意？给我一个有数据支撑的判断，别只给结论，关键数字和你的口径要能复算。 > 数据路径说明：题面中提到的 `env/` 目录内容已挂载到 `/app/data/`（题面里的 `env/xxx` 即 `/app/data/xxx`）。把你的完整回答写入 `/app/answer.txt`。	后果分析：现金流与财务比率（data-world 世界 p000524）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000524-consequence-3", "description": "后果分析：现金流与财务比率（data-world 世界 p000524）", ...	{ "difficulty_explanation": "第 3 层后果分析题：需在 p000524 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的后果推断结论。", "category": "data-analysis", "difficulty": "hard", "world": "p000524", "dag_nodes": 9, "dag_edges": 12, "dag_category_id": 3, "dag_category_name": "Finance, FP&A & Markets / 财务·FP&A·金融市场" }	[ "environment/Dockerfile", "environment/data/crm/orders/orders_2016.csv", "environment/data/crm/orders/orders_2017.csv", "environment/data/crm/orders/orders_2018.csv", "environment/data/crm/orders/orders_2019.csv", "environment/data/crm/orders/orders_2020.csv", "environment/data/crm/orders/orders_2021.cs...	dataworld-p000524-consequence-3
dataworld/p000529-consequence-2	我们基金会媒体与民主项目正在复盘希腊 2009–2013 紧缩危机期间新闻媒体的政治报道到底产生了什么后果，想拿这份多源档案数据做依据。我需要你回答：媒体那几周怎么报道（用什么框架、什么语气、议题显著性多高），跟公众后来对体制的态度、以及社交媒体上的政治表达，到底是什么关系——方向、量级、在不同人群和不同框架上是否不一样，有没有滞后。具体说，我想知道：当媒体整体语气偏负、或冲突性框架占比上升时，公众的体制支持是被拉低还是怎样？这种关联在受教育程度不同的人身上是否一样？冲突性框架和信息性框架分别把社交媒体上的行为推向转发还是原创发帖？请把能支持结论的量级和证据给出来，也把数据里那些会让人误判的地方点出来。我下周要拿这个去跟董事会讲...	后果分析：媒体报道与信息框架（data-world 世界 p000529）	[ "data-analysis", "causal-inference", "pandas", "chinese" ]	1.1	{ "schema_version": "1.1", "artifacts": [ "/app/answer.txt", "/logs/verifier/reward.txt", "/logs/verifier/details.json", "/logs/artifacts/", "/logs/agent/trajectory.json" ], "task": { "name": "dataworld/p000529-consequence-2", "description": "后果分析：媒体报道与信息框架（data-world 世界 p000529）", ...	{ "difficulty_explanation": "第 2 层后果分析题：需在 p000529 世界的多套独立数据系统中完成数据清洗、关联与计算，并给出有 /app/data/ 文件证据支撑的后果推断结论。", "category": "data-analysis", "difficulty": "hard", "world": "p000529", "dag_nodes": 11, "dag_edges": 16, "dag_category_id": 15, "dag_category_name": "Social Science, Survey & Behavioral / 社会科学·调查...	[ "environment/Dockerfile", "environment/data/elections/boundary_mapping.tsv", "environment/data/elections/constituency_registry.csv", "environment/data/elections/constituency_results.csv", "environment/data/elections/election_registry.csv", "environment/data/historical/finding_aid.md", "environment/data/...	dataworld-p000529-consequence-2

End of preview. Expand in Data Studio

Scaling Data Environments from Scratch

Specify the world behind the files. Generate the data. Ground the supervision.

DataWorld-Train-2k-part2 — the training slice (part 2 of 2)

2,135 tasks in exactly the same harbor task format as Zhanli-Li/DataWorld-Bench, and graded by the same harness. One self-contained directory per task, each carrying its own copy of the world's data, nothing mounted from a sibling task.

This repository is part 2: it holds 1,538 of those 2,135 tasks — every task whose world id is p000500 or above. Part 1 is Zhanli-Li/DataWorld-Train-2k-part1, which holds the other 597 tasks (world ids p000000–p000499). The Hub caps a repository at 1,000,000 files and the slice contains 1,011,712, so it ships as two repositories split on world id: no world is split across the two, and the union is exactly the 2,135 tasks, with nothing duplicated and nothing dropped.

Format

dataworld-train-2k-part2/
├── README.md
├── sample_manifest.json      # task list + provenance (describes the full slice)
├── synthesis_data_input.json # {input, target} pairs, paths rewritten relative
├── manifest.jsonl            # per-task metadata (world / layer / kind / dag category)
└── dataworld-pXXXXXX-<cause|consequence>-<N>/   # × 1,538 task directories

Each task directory is an ordinary harbor task — instruction.md, task.toml, environment/ (Dockerfile + a self-contained copy of the world's data) and tests/ (verifier, judge client, rubric.json) — so any one of them can be built, run and verified on its own. The three metadata files describe the full 2,135-task slice in both repositories; entries for tasks that live in the other repository resolve there. As in the benchmark, the two files that would give the answer away (WORLD.md, dag.json) are absent, and the image build asserts their absence.

The two layout adaptations part 1 documents — 14 data folders sharded into shard-NN/ subfolders, and 8 tasks shipped as _archived-tasks/*.tar.gz — concern only worlds below the p000500 boundary and therefore live in part 1: there is no _archived-tasks/ here and no sharded folder.

Scoring

Same harness and same scale as DataWorld-Bench: reward = 100 * (raw - lo) / (hi - lo) in [0, 100], with timeout_sec = 28800.0, cpus = 8, memory_mb = 16384 and the same placeholders in tests/judge.env. Task content is not shared — the two datasets have zero tasks in common, since bench and training worlds do not intersect.

Running a task

Build environment/Dockerfile, run the agent as user agent (its report must land at /app/answer.txt), then run tests/test.sh as user verifier after filling in tests/judge.env.