Harvey LAB

Harvey LAB is an open-source benchmark for evaluating how effectively AI agents perform realistic legal work. It combines a dataset of legal assignments with an execution harness that runs agents against those tasks. Each benchmark task includes instructions, working documents, and rubrics that define successful completion. The environment is designed to test practical legal workflows rather than isolated question-answering ability. Evaluation tools score agent outputs and support reports and comparative experiment runs. Documentation includes an end-to-end M&A data-room example covering setup, execution, scoring, and result analysis. The project is intended to help researchers and developers identify weaknesses and measure improvements in legal AI agents.