Decision-Relative Authority Closure: When Information Exists but Cannot Be Permissibly Acquired to Resolve a Decision

Modern autonomous systems can acquire additional evidence when a decision is uncertain, but a prior question is often left implicit: is the system authorized to perform the experiment needed to resolve that uncertainty? We develop Decision-Relative Authority Closure (DRAC), a finite framework for adaptive evidence acquisition when experiment admissibility depends on the still-unresolved decision class. The model separates technical observability from authorization-resolvability and captures cases in which incompatible worlds are distinguishable in principle but no admissible sequence reaches the required evidence. We prove hereditary admissibility and an Authorization-Closure Characterization Theorem: under the deterministic hereditary rule, an adaptive admissible experiment tree resolves the required decision if and only if every block of the authorization-closure fixed point is decision-homogeneous. The result yields an order-independent feasibility criterion and a direct fixed-point algorithm. We then implement the theory and verify it against an independently coded recursive adaptive-policy solver. Across 19,260 exhaustively generated binary systems and 25,000 higher-dimensional randomized systems—44,260 systems in total—the procedures disagree in zero cases. Moreover, 16,676 exhaustive systems contain decision-separating signal while remaining authorization-unresolvable. The framework therefore identifies a precise gap between information that exists and information that can be permissibly acquired for action.