Under Which Mandate? Representing the Institutional Provenance of Agent Authority

Agent platforms are converging on runtime identity, scoped permissions, policy gates and audit trails. These establish which agent acted and whether a tool call was technically permitted. They do not establish whether the action rested on an institutionally grounded decision right: which governance body granted it, under which mandate, within which autonomy boundary, and who remains answerable. This paper calls the missing machine-readable connection between technical authority to execute and the institutional constitution of authority to decide the institutional semantics gap. It is operationalised as a design-science artefact: a decision-centred ontology built on the chain Institution → Governance Body → Mandate → Decision Right → Decision → Governed Object, with typed decision rights, an Ecore metamodel constrained by twenty-one OCL invariants, and a check service that an agent's tool gate consults before every state-changing call. A demonstration against a deliberately simple baseline, in which two agents driven by a locally run language model act in seventeen scenarios, shows the chain enforced in front of a real agent: actions that a baseline of identity, permission and some human approval executes and the institutional check stops; one case that no pre-execution gate could see, because the decision record that reveals it existed only afterwards; an agent that tried a second tool after being refused; and a defect in the artefact, an approval not bound to the action it approved, corrected and shown closed in a second run. Three propositions on occurrence, effect and detection, and the designs to test them, close the paper.