Liberal democracies face a monitoring deficit: voters cannot observe administrative behavior between elections, and existing oversight institutions are either partisan principals lacking epistemic independence or advisory bodies lacking investigative authority. This paper proposes the Epistemic Censorate, a sortition-based citizen audit body with subpoena power and public reporting requirements, modeled on imperial China's Censorate and contemporary deliberative institutions. Drawing on principal-agent theory, democratic theory, and comparative institutional analysis, I develop a theoretical model in which a randomly selected civic principal reduces information asymmetry between state and society, creates continuous accountability, and complements existing oversight mechanisms. I specify institutional design features, derive testable predictions, address objections, and outline a pilot research design. The paper contributes a concrete, theoretically grounded institutional innovation to debates on democratic reform.