Temporal Integrity in Explainable AI: A Post-hoc Audit Perspective Based on Temporal Boundary Violation

This work presents a conceptual examination of temporal integrity in AI-generated explanations, focusing on whether an explanation could have been constructed using only information available at the original decision time (t₀). The paper introduces Temporal Boundary Violation (TBV) as a descriptive concept for identifying structural risks in post-hoc explanations, where knowledge acquired after a decision is retrospectively incorporated into accounts that purport to reflect earlier reasoning. TBV is distinguished from prediction error, factual inaccuracy, or bias, and is instead framed as a deviation in the temporal organization of explanations. The scope of this work is intentionally limited to conceptual framing and problem delineation. It does not propose formal definitions, evaluation metrics, decision criteria, implementation procedures, or domain-specific applications. The objective is to clarify an underexamined dimension of explainability and to situate TBV as a post-hoc audit perspective that complements existing Explainable AI (XAI) research. This record serves as a conceptual registration and reference point. Empirical validation, quantitative indicators, operational protocols, and commercial applications are intentionally deferred to separate work. This work is part of the ongoing 'Polymorphic Boundary Case Studies' by TBV Corporate (CTIN Institute).