LongDA is a data analysis benchmark for evaluating LLM-based agents under documentation-intensive analytical workflows. It features authentic U.S. government survey data with complete, long documentation, testing LLMs' ability to navigate complex real-world datasets before performing analysis.
