datasets/task_metadata.json

{
  "easy": {
    "task_id": "easy_dedup_rename",
    "difficulty": "easy",
    "description": "Clean an employee dataset by: 1) Removing duplicate rows, 2) Renaming columns to snake_case format. Call finish when done.",
    "max_steps": 10,
    "operations_allowed": [
      "remove_duplicates",
      "rename_columns",
      "finish"
    ],
    "scoring": {
      "duplicate_score": 0.5,
      "schema_score": 0.5
    }
  },
  "medium": {
    "task_id": "medium_missing_dtype",
    "difficulty": "medium",
    "description": "Clean a customer dataset by: 1) Filling missing values using correct strategy (mean for numeric, mode for categorical), 2) Fixing data types (age, purchases, salary must be numeric). Call finish when done.",
    "max_steps": 15,
    "operations_allowed": [
      "fill_missing_mean",
      "fill_missing_mode",
      "fill_missing_median",
      "fix_dtype",
      "finish"
    ],
    "scoring": {
      "missing_score": 0.5,
      "dtype_score": 0.5
    }
  },
  "hard": {
    "task_id": "hard_full_pipeline",
    "difficulty": "hard",
    "description": "Clean an orders dataset by running a full pipeline: 1) Remove duplicate orders, 2) Fill missing values, 3) Fix data types, 4) Remove outliers in quantity and price columns, 5) Validate final schema. Call finish when done.",
    "max_steps": 20,
    "operations_allowed": [
      "remove_duplicates",
      "fill_missing_mean",
      "fill_missing_mode",
      "fill_missing_median",
      "fix_dtype",
      "remove_outliers",
      "validate_schema",
      "finish"
    ],
    "scoring": {
      "duplicate_score": 0.2,
      "missing_score": 0.2,
      "dtype_score": 0.2,
      "outlier_score": 0.2,
      "schema_score": 0.2
    }
  },
  "expert": {
    "task_id": "expert_sales_pipeline",
    "difficulty": "expert",
    "description": "Clean a sales dataset by running expert pipeline in ORDER: 1) Remove duplicate transactions, 2) Rename columns to snake_case, 3) Fill missing values using mode strategy, 4) Fix all data types, 5) Remove outliers using IQR method, 6) Validate final schema. Call finish when done.",
    "max_steps": 25,
    "operations_allowed": [
      "remove_duplicates",
      "rename_columns",
      "fill_missing_mean",
      "fill_missing_mode",
      "fill_missing_median",
      "fix_dtype",
      "remove_outliers",
      "validate_schema",
      "finish"
    ],
    "scoring": {
      "duplicate_score": 0.15,
      "missing_score": 0.2,
      "dtype_score": 0.2,
      "outlier_score": 0.2,
      "schema_score": 0.25
    }
  }
}