The Mandate as a Unit of Risk — paper outline

This research outline sets out the proposed argument, derivations, evidence requirements and limitations for “The Mandate as a Unit of Risk.” The proposed paper connects the authority under which an AI agent acts with the economics of task reliability, oversight and adverse outcomes. It treats the mandate as a versioned authority for a class of acts, enforced through a permission that can be refused before an act is executed. The research programme examines how a mandate can express an admissible set of model-and-oversight configurations and a quantitative risk appetite, with authority narrowing automatically on independently adjudicated evidence and widening only with an authorised human signature. It extends the task-value framework with the cost of refusal and escalation, examines calibration in financial services, and identifies the conditions and limitations of any proposed assurance statement. The outline distinguishes established precedents from candidate contributions and identifies simulation and empirical work still required. Publication status: This deposit is a research outline, not a completed or peer-reviewed paper. Proposed theoretical results, calibration procedures and empirical tests remain to be established. Cordier’s simulation layer has not been independently rerun; the outline states the assumptions on which its proposed analysis depends.