Why Some Actions Cannot Be Chosen: Authorization Preceding Optimization in Long-Horizon Systems

This work introduces a negative architectural result for adaptive and agent-based systems: some actions cannot be explained as the result of choice or optimization. The paper isolates a structural distinction between authorization and choice, showing that for actions with irreversible consequences, admissibility must be resolved prior to any act of deliberation or comparison. As a result, action execution implies prior authorization, independently of utility, reward, belief, or preference. The contribution is intentionally minimal and architectural. No controllers, learning algorithms, or implementation details are provided. The goal is to establish the existence and necessity of an authorization layer that precedes optimization in long-horizon systems, and to define a boundary condition that any viable agent architecture must respect. This document serves as both prior art and an anchor for a broader class of control and cybernetic architectures that operate beyond standard optimization-based frameworks, opening the door to Cybernetics 2.5.