Assessment method that examines not just accuracy but also what evidence or features a model actually uses to make decisions.