BoolQ consists of genuine questions posed by users (often from search queries) paired with relevant passages. Models must decide whether the evidence entails a “yes” or “no” answer, making the task sensitive to careful reading, negation handling, and calibration on borderline cases.

Despite the apparent simplicity of binary answers, high performance requires precise comprehension of the passage and question intent. BoolQ is commonly included as a compact probe of factual reading and robustness to linguistic nuance.