Simulation Is Not Adjudication Institutional Admissibility and the Boundary Between Machine Output and Legal Consequence

AI systems increasingly produce outputs that emulate the form of legal interpretation, adjudicative reasoning, and compliance assessment.

Existing governance frameworks address accuracy, explainability, and auditability, but leave under-articulated a prior question: whether a machine output can cross the boundary from technical event to institutional consequence.

This paper argues, through conceptual analysis drawing on philosophy of mind (Seth, 2021; Searle, 1980), legal theory (Hart, 1961; Raz, 1979; Alchourrón and Bulygin, 1971), and AI governance (EU AI Act, Reg. 2024/1689), that simulation of legal decision-making is not institutional legal realization. It identifies four category errors that produce governance failures: treating optimization as interpretation, traceability as admissibility, local validity as aggregate legitimacy, and technical execution as institutional consequence.

Drawing on the tradition of legal logic from Huerta Ochoa (2025), Bulygin (2015), Carrió (1965), Bobbio (1960), and Nino (1985), the paper argues that legal logic is not computational formalism: the distinction between norms, normative statements, and normative propositions shows that machine outputs, however logically coherent, occupy a different plane from the institutional act of norm application.

The paper then proposes institutional runtime admissibility as a governance category positioned against existing doctrines of evidential admissibility, Hartian validity, and Razian authority, filling the gap none of them covers: the continuing condition that must remain valid at the moment a machine output is about to produce institutional consequence.