Institutions operating under AI-mediated decision processes face a structural problem that existing frameworks do not fully resolve: the conditions under which consequence can legitimately attach to a decision pathway are not equivalent to the conditions under which an AI output appears stable, coherent or actionable. This document defines Decision-Pathway Admissibility as the structural boundary condition under which consequence can legitimately attach to an AI-mediated decision pathway in institutional contexts. It distinguishes between output stability (the coherence or consistency of a model output) and consequence-bound stability (the structural validity of the decision pathway through which institutional consequence is assigned). These are not equivalent conditions. Treating them as equivalent creates a structural gap with direct implications for liability routing, insurability and audit coherence. A decision pathway is admissible only when four conditions are jointly satisfied: attributability, reconstructability, interruptibility and evidential grounding. These conditions constitute a single structural precondition, not a checklist derived from existing governance or compliance frameworks, under which consequence can attach and existing institutional mechanisms can operate coherently. This is a non-prescriptive classification statement. It does not define implementation, compliance or governance mechanisms. It defines the boundary condition upon which such mechanisms depend.
Victoria Gavaza (Mon,) studied this question.