Using a tool-using agent to systematically explore and identify data quality issues in a corpus by compiling findings into executable tests.