Enterprises increasingly route requests among models, providers, endpoints, service tiers, compute budgets, and fallback paths at runtime. Technical research has established that such selection works. It typically takes as given the things a firm actually has to decide: which configurations are eligible, what the routing policy is optimizing, what evidence justifies it, and who may change it. This paper takes up that management problem. It introduces the Model Choice Boundary, the allocation of decision rights between an enterprise and the intermediaries that route its inference, and separates the sourcing boundary, who supplies or operates the routing mechanism, from the authority boundary, who controls the decision itself. Model choice is decomposed into rights over eligibility, objective, change, allocation, intervention, evidence, learning, and exit, allocated through workload specific routing mandates and verified through a four level assurance architecture. The central claim is that a firm can delegate high frequency allocation without surrendering strategic control, but only where retained rights are backed by the evidence, evaluation, intervention, and exit capabilities that make authority real rather than nominal. One corollary is immediately checkable: a routing mandate that names several goals without ordering them has delegated the objective, whatever the agreement says. And static assignment remains a legitimate outcome wherever the incremental value of delegation does not cover its economic, control, and assurance costs.
This is a living draft. Versions are numbered and dated on the title line. Comments and criticism are welcome: contact@jamesemurphy.com.
James E. Murphy, “Model Choice Boundary: Sourcing, Delegation, and Assurance for Runtime Model Selection in Enterprise AI,” working draft v1.0, 14 September 2026. Available at https://jamesemurphy.com/ai-economics/model-routing/.