SCORING METHODS

EXACT-MATCH SCORING Exact-match scoring checks whether the agent's output is identical to the expected answer. If the expected answer is 42 and the agent outputs 42, the score is a pass. If the agent outputs "The answer is 42," a strict exact match fails it, because the text is not identical. Exact match is fast, free, perfectly consistent, and easy to explain. It is excellent for questions with a single short answer, such as numbers, yes or no, a multiple-choice letter, or a specific code. Its weakness is brittleness. Brittle means easily broken, like a dry twig. Tiny harmless differences, such as extra spaces, capital letters, or a trailing period, can cause failure unless you first normalize the text. Normalizing means cleaning both answers into a standard form, such as lowercase text with extra spaces removed, before comparing.