Evals

Short for evaluations. Test suites that measure how well a model or AI feature performs on realistic tasks, so you can tell whether a change made it better or worse.

“Before switching models, run the evals on our real customer questions.”