Preserving Human Oversight Through Bounded Explanation Deferral in Intelligent Systems

Abstract Adaptive interfaces can reduce immediate information demands by postponing explanations, yet postponed material may remain essential for subsequent human oversight. This paper develops a theoretical framework for bounded explanation deferral in human–intelligent systems integration. It distinguishes an invariant decision-critical information layer from deferrable supporting material and treats the latter as a review obligation with an owner and deadline. A deterministic capacity balance connects present information demand, accumulated explanation debt, recovery opportunities and uncertainty about available attention. Analytical reasoning establishes a necessary cumulative feasibility condition and shows why deferral alone cannot resolve persistent demand. A transparent six-window worked example illustrates how scheduled repayment can remove temporary capacity exceedances while clearing deferred material before a dependent decision. Counterexamples demonstrate failure under insufficient recovery capacity, underestimated reading effort and misclassified critical evidence. These are mathematical illustrations, not participant observations or validated performance effects. The contribution is an auditable specification for adapting information timing while preserving contestability and identifying when operational escalation is required. The framework yields falsifiable propositions for later human-subject evaluation without assuming that lower workload necessarily improves safety.