yjijn commited on
Commit
6b586af
·
1 Parent(s): 0fdffdb

Add database_clear as a denoised control environment

Browse files

Provide aligned, noise-removed counterparts of assets/database so evaluations can isolate how data friction affects measured model capability.

README.md CHANGED
@@ -71,7 +71,9 @@ DataClawBench simulates at scale the noisy, weakly-semantic, cross-domain data e
71
 
72
  Key directories and scripts:
73
 
74
- - `assets/database/`: Benchmark data files, injected wholesale into the container workspace at run time. The root contains `internal_metrics.csv` (internal business-logic knowledge base); `enterprise/`, `industry/`, and `policy/` hold the three theme-domain datasets.
 
 
75
  - `assets/qa_raw/`: Raw task source files.
76
  - `assets/qa_gold/`: Minimized gold files derived from `qa_raw`.
77
  - `tasks/`: Generated OpenClaw task spec files.
@@ -362,6 +364,8 @@ Under a theme-domain view and business-oriented taxonomy, the current data envir
362
 
363
  > At execution time, agents typically need to align entities across files, join across tables, normalize definitions, and perform aggregation calculations, rather than simply looking up values in a single file; when needed, they must also consult business conventions in `internal_metrics.csv`. This is the core value of DataClawBench for evaluating real-scenario data understanding and reasoning capabilities.
364
 
 
 
365
  ### 📋 Task Statistics
366
 
367
  The current version contains **492** tasks across **7** categories, with an overall difficulty distribution of **131 easy / 286 medium / 75 hard**.
 
71
 
72
  Key directories and scripts:
73
 
74
+ - `assets/database/`: Benchmark data files, injected wholesale into the container workspace at run time. The root contains
75
+ `internal_metrics.csv` (internal business-logic knowledge base); `enterprise/`, `industry/`, and `policy/` hold the three theme-domain datasets.
76
+ - `assets/database_clear/`: Cleaned counterpart of `assets/database/`. Built from the original data after completing data alignment and removing noise; provided as a control environment for measuring how the data environment affects model capability.
77
  - `assets/qa_raw/`: Raw task source files.
78
  - `assets/qa_gold/`: Minimized gold files derived from `qa_raw`.
79
  - `tasks/`: Generated OpenClaw task spec files.
 
364
 
365
  > At execution time, agents typically need to align entities across files, join across tables, normalize definitions, and perform aggregation calculations, rather than simply looking up values in a single file; when needed, they must also consult business conventions in `internal_metrics.csv`. This is the core value of DataClawBench for evaluating real-scenario data understanding and reasoning capabilities.
366
 
367
+ **Cleaned control environment (`assets/database_clear/`).** In addition to the default noisy environment above, this repository provides `assets/database_clear/`: a cleaned counterpart of the same sources after data alignment and noise removal. Comparing evaluation results on `database` vs. `database_clear` helps isolate how much of the observed performance is attributable to environmental friction rather than to the model itself.
368
+
369
  ### 📋 Task Statistics
370
 
371
  The current version contains **492** tasks across **7** categories, with an overall difficulty distribution of **131 easy / 286 medium / 75 hard**.
README.zh-CN.md CHANGED
@@ -48,6 +48,7 @@ DataClawBench 大规模地模拟了高噪声、弱语义、跨领域的真实世
48
  关键目录与脚本职责如下:
49
 
50
  - `assets/database/`:评测数据文件,运行时会整体注入到容器工作区。根目录含 `internal_metrics.csv`(业务逻辑内部知识库);`enterprise/`、`industry/`、`policy/` 为三大主题域数据。
 
51
  - `assets/qa_raw/`:原始任务源文件。
52
  - `assets/qa_gold/`:由 `qa_raw` 归约得到的精简 gold 文件。
53
  - `tasks/`:生成后的 OpenClaw task 规范文件。
@@ -338,6 +339,9 @@ DataClawBench 数据并非来源于合成样本或教学示例,而是基于发
338
 
339
  > 在任务执行层面,智能体通常需要在多文件间完成实体对齐、跨表关联、口径归一与聚合计算,而非单文件查值;必要时还需结合 `internal_metrics.csv` 中的业务约定。这也是 DataClawBench 用于评估真实场景数据理解与推理能力的核心价值。
340
 
 
 
 
341
  ### 📋 任务统计信息
342
 
343
  当前版本共 492 个任务,覆盖 7 个类别;整体难度分布为 131 easy / 286 medium / 75 hard。
 
48
  关键目录与脚本职责如下:
49
 
50
  - `assets/database/`:评测数据文件,运行时会整体注入到容器工作区。根目录含 `internal_metrics.csv`(业务逻辑内部知识库);`enterprise/`、`industry/`、`policy/` 为三大主题域数据。
51
+ - `assets/database_clear/`:在 `assets/database/` 基础上完成数据对齐、去除数据噪声后的干净数据,用作对照环境,便于测定数据环境对模型能力的影响。
52
  - `assets/qa_raw/`:原始任务源文件。
53
  - `assets/qa_gold/`:由 `qa_raw` 归约得到的精简 gold 文件。
54
  - `tasks/`:生成后的 OpenClaw task 规范文件。
 
339
 
340
  > 在任务执行层面,智能体通常需要在多文件间完成实体对齐、跨表关联、口径归一与聚合计算,而非单文件查值;必要时还需结合 `internal_metrics.csv` 中的业务约定。这也是 DataClawBench 用于评估真实场景数据理解与推理能力的核心价值。
341
 
342
+ **干净对照环境(`assets/database_clear/`)。** 除上述默认的高噪声数据环境外,本仓库还提供 `assets/database_clear/`:在原数据基础上完成数据对齐并去除噪声后的干净数据。对比 `database` 与 `database_clear` 两套环境上的评测结果,便于测定数据环境对模型能力的影响,将环境摩擦与模型能力本身区分开来。
343
+
344
+
345
  ### 📋 任务统计信息
346
 
347
  当前版本共 492 个任务,覆盖 7 个类别;整体难度分布为 131 easy / 286 medium / 75 hard。
assets/database_clear/enterprise/company_core.csv ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d5dca195a01a7b368fde4af6f8284c34f723147bcaf8e3527bfe02aecf718e82
3
+ size 18714586
assets/database_clear/enterprise/company_operation_status.csv ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4af094ff21fba8c4d856de7a0b2c3159d0af0a4bd70cc36b4faff0930e75486b
3
+ size 1781476
assets/database_clear/enterprise/company_profile.csv ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ecd68737f4f453746c3271b146ef5f6b6e741536aab4fc8ae2a32be89085e3c2
3
+ size 1783742
assets/database_clear/enterprise/company_translation_english_chinese.json ADDED
The diff for this file is too large to render. See raw diff
 
assets/database_clear/industry/national_industry_status.csv ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fc0853e0ecde4d4fe73d128d79fbdc7930bed9b57cfc718c68f554c2cea7517a
3
+ size 79920
assets/database_clear/industry/regional_industry_status.csv ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ad07d2990eea08b4aaa914588f7856461ea3c3de289b72176501756a937e079b
3
+ size 1013131
assets/database_clear/internal_metrics.csv ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f584fb7cacf19fcdc50e59b398fc65ad041d9651860d9cfe1d7afb277dbeb9dc
3
+ size 2238
assets/database_clear/policy/policy_release_status.csv ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a784a9301e6a8feb63d7b5e2fae7da0f4a3fee3d1871703cf1560d36163184ef
3
+ size 1109101
assets/database_clear/policy/policy_resource.csv ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8322256232d053b49b7b0e7f842e811224a203f07bf51d1c890c62bdb6ea1a6f
3
+ size 20833538