The degree to which an annotator (human or model) produces consistent judgments, typically measured by agreement metrics like F1 score.