DataClawEval Benchmark for Data Engineering Agents

DataClawEval introduces 100 end-to-end tasks to evaluate autonomous agents across PySpark, MySQL, HiveSQL, PrestoSQL, and FlinkSQL. Unlike LLM-as-a-judge methods, the benchmark uses deterministic grading within isolated sandboxes to assess real-world industrial data engineering capabilities.