Machine learning-based credit risk models are commonly assessed using discrimination metrics alone. Such evaluation, however, does not fully capture economic consequences, temporal deployment conditions, or changes in the underlying risk environment. This study develops a decision-aligned,...