A knowledge graph separates knowing that a relation exists from learning how strongly evidence supports it. This distinction matters because inference can be inspected as a path or rule, while machine learning often compresses justification into parameters. The hybrid system therefore raises an epistemic question: is an explanation a faithful account of the computation, or merely a persuasive reconstruction? Graph structure improves legibility, but semantic labels and ontology choices still encode human judgments about what counts as an entity, cause, or relevant connection.