Despite that independent agencies are typically justified in terms of technical efficiency, they inevitably have to make political judgments. How can political reasoning be legitimate in such institutions? This paper starts by investigating the merits of two prominent models. The “avoidance model” asks agency reasoning to stick to empirical facts and as far as possible stay clear of political values. By contrast, the “specification model” recognizes the need for constructive normative work, but confines it to the refinement of given statutes. This paper challenges both models and defends a third alternative. The “public reason model” requires agencies to ground their value judgments in a publicly accessible framework of reasoning, which is here interpreted as their overarching mandate. The paper argues that agency mandates should be conceived as distinct domains of reasoning, and it delineates three institutional virtues that enable agencies to track this domain.