<p>Administrative AI systems increasingly support document processing and regulatory workflows in document-intensive domains such as foreign food facility registration. However, in administrative environments where legal responsibility and interpretive discretion remain inseparable from judgment, automation-centered approaches risk obscuring accountability and judgment attribution. This study proposes a reasoning representation framework for accountability-preserving AI that structures human judgment within administrative decision support systems. The framework introduces the Judgment Reason Code (JRC), a hierarchical domain–subtype–tag structure designed to externalize the interpretive logic underlying administrative decisions without automating decision authority. Rather than predicting outcomes, the framework encodes the reasoning basis of judgments as reusable, machine-interpretable units. Using scenario-based structural validation, the study examines whether administrative reasoning can be consistently represented, differentiated across distinct judgment contexts, and retrieved through similarity-based linkage. The evaluation focuses on representability, distinguishability, retrievability, and non-automation integrity rather than predictive performance metrics. By separating reasoning representation from decision execution, the proposed framework enables the accumulation and retrieval of institutional reasoning across cases while preserving explicit human decision authority. The study thus establishes a foundational design perspective for reasoning-oriented AI systems in administrative decision support contexts.</p>