A research-backed framework for evaluating LLM-generated data transformations. Learn how datasets, sandboxed execution, and automated judging reveal failure patterns and model performance across real-world data engineering tasks.