Precision measures how often a classifier’s positive predictions are correct: among all positive predictions, it counts the share that match the intended positive label. In language-model evaluation, it helps assess whether an automated judge agrees with expert annotations. High precision means positive judgments are usually trustworthy, though precision alone does not show how many true positives were missed. section 7.10
For example, an LLM judge can label real-estate descriptions for enthusiasm or positivity and then be compared with human labels. Precision and recall are preferable to relying only on overall accuracy when positive and negative examples are unevenly represented. section 7.10
