Public Auditing of Explanation: Stopping Points, Explanatory Graphs, and the Impossibility of Coronation

This paper develops a framework for the public auditing of submitted explanations. What an explainer submits is not an explanatory graph but an explanation in raw form; the graph is the object reconstructed by an interpreter under publicly reviewable procedures. I begin by distinguishing the explainer from the interpreter and, on that basis, fixing three methodological stopping points: explanations are publicly reconstructible, explanatory commitments can be identified, and compression can be distinguished from decompression. I then argue that any submitted explanation can, after audit, be reconstructed as an explanatory graph with a reconstruction layer and an audit layer. The audit layer can always be rendered as a finite directed acyclic graph, so frontier nodes must exist. I further define explanatory vindication in a strict sense as the deterministic transmission of a node’s standing from a non-empty upstream base set through deductive decompression traceable within the graph. It follows that no frontier stopping point can be explanatorily vindicated within the graph. It can only be defended.