AI evaluations are structured tests that measure whether a model or system demonstrates defined capabilities, limitations, safety properties, and operational performance. This guide explains the mechanism, trade-offs, evaluation, and controls that matter in practice.