This paper does not propose a standard, recommend adoption, or evaluate implementations. Its purpose is to define the class of artifact required, in principle, if AI-mediated reasoning is to remain explainable under later review.