A framework for evaluating AI by explanation, contestability, accessibility, and fit—not just accuracy.