An independent fact-checking layer for AI output that returns a sourced verdict for any factual claim, either a fast check or the full eight-model trace.