Cohort Slicing

An AI evaluation method that partitions test data into groups along specific dimensions—such as language, region, or user type—and evaluates each group independently, in order to surface localized failure patterns that aggregate metrics may obscure.