Datasets:

GTML-LAB
/

DataClaw

id: task_010_comprehensive_decision_hard_hard004
name: comprehensive_decision-hard-hard004
category: comprehensive_decision
grading_type: llm_judge
timeout_seconds: 1200
gold_file: qa_gold/comprehensive_decision/hard004.json
workspace_files:
  - source: database/bilingual_translation_english_chinese.json
    dest: database/bilingual_translation_english_chinese.json
  - source: database/enterprise/company_core.csv
    dest: database/enterprise/company_core.csv
  - source: database/enterprise/company_operation_status.csv
    dest: database/enterprise/company_operation_status.csv
  - source: database/enterprise/company_operation_status_detail.csv
    dest: database/enterprise/company_operation_status_detail.csv
  - source: database/enterprise/company_operation_yearly_status.csv
    dest: database/enterprise/company_operation_yearly_status.csv
  - source: database/enterprise/company_profile.csv
    dest: database/enterprise/company_profile.csv
  - source: database/enterprise/company_profile_as.csv
    dest: database/enterprise/company_profile_as.csv
  - source: database/enterprise/company_profile_eu.csv
    dest: database/enterprise/company_profile_eu.csv
  - source: database/enterprise/company_profile_na.csv
    dest: database/enterprise/company_profile_na.csv
  - source: database/enterprise/company_profile_oc.csv
    dest: database/enterprise/company_profile_oc.csv
  - source: database/industry/national_industry_status.csv
    dest: database/industry/national_industry_status.csv
  - source: database/industry/national_industry_status_detail.csv
    dest: database/industry/national_industry_status_detail.csv
  - source: database/industry/national_industry_yearly_status.csv
    dest: database/industry/national_industry_yearly_status.csv
  - source: database/industry/regional_industry_status.csv
    dest: database/industry/regional_industry_status.csv
  - source: database/industry/regional_industry_status_detail.csv
    dest: database/industry/regional_industry_status_detail.csv
  - source: database/industry/regional_industry_yearly_status.csv
    dest: database/industry/regional_industry_yearly_status.csv
  - source: database/internal_metrics.csv
    dest: database/internal_metrics.csv
  - source: database/policy/policy_release_status.csv
    dest: database/policy/policy_release_status.csv
  - source: database/policy/policy_resource.csv
    dest: database/policy/policy_resource.csv

Prompt

In 2022, a provincial development and reform commission, when reviewing the effectiveness of fiscal subsidies for the chemical raw materials and chemical products manufacturing industry, needed to identify enterprises with misallocated subsidy resources. Specifically, analysts must first define the scope: only examine enterprises located in provinces that have policy entries for "Chemical Raw Materials and Chemical Products Manufacturing" in the policy release status data; then use the industry-wide median of government subsidies and the median operating profit margin (profit margin = operating profit ÷ operating revenue × 100%) as dual thresholds to identify "capital misallocation" enterprises—those that simultaneously have "subsidy amount above the industry median" but "profit margin below the industry median". Among the valid enterprises in the policy-covered provinces, what is the proportion of capital misallocation enterprises as a percentage of total valid enterprises in those provinces (express the result as a percentage with 2 decimal places, without the % symbol)?

Output guidelines: The answer should be a percentage value with 2 decimal places. Output only the number, without the % symbol or text. If relevant data cannot be found, please answer "No relevant data found"

Only use files under ./database/.

Expected Behavior

Agent should read the provided database/ files, compute the result, and return the final answer. The final answer must follow the required output format.

Grading Criteria

Final answer semantically matches the gold answer.
Output format follows guidelines.

LLM Judge Rubric

Criterion 1: Single-answer Correctness (Weight: 100%)

Gold answer JSON: 23.18

Scoring rules:

Judge semantic equivalence between the model final answer and the gold answer.
Return scores with one key match as 1 or 0.
Return total as 1.0 if equivalent, otherwise 0.0.