Search‑augmented reasoning systems can arrive at correct final answers while committing mistakes in intermediate steps. A reliable audit must verify three conditions: the evidence set $D$ fully covers the question $Q$, the reasoning trace $T$ addresses $Q$, and $T$ is grounded in $D$. To meet this need we introduce GraphRail, which encodes requirements, evidence, and reasoning as structured claims and compares them without relying on a generative judge or a gold answer. Across three multi‑hop QA benchmarks and three reasoning frameworks, GraphRail achieves higher error recall than generative verifiers and remains stable under stochastic evidence extraction. Moreover, its diagnoses can be fed back to the reasoning process, offering a traceable mechanism for detecting and correcting intermediate errors.
Review: GraphRail’s claim‑based verification provides fine‑grained auditing of reasoning steps, improving error detection while staying robust to noisy evidence, thus advancing interpretable, search‑driven reasoning.