Legal Judgment Prediction (LJP) is a core task in the field of legal artificial intelligence, with the primary objective of predicting legal judgment outcomes based on detailed case descriptions. However, how to make these predictions both interpretable and grounded in legal reasoning remains a key challenge. This paper proposes an innovative binary evaluation model that uses a series of binary judgments (e.g., legal/illegal, justified/unjustified) to enhance legal reasoning tasks. Guided by a multi-stage prompting framework, the model effectively enhances interpretability. Experimental results on the CAIL2018 dataset demonstrate its significant contributions to transparency and consistency, offering a feasible solution to the‘black box’ problem.