The Introduction states that the paper's novelty is methodological rather than algorithmic: strict-parity graph-RAG evaluation, leakage auditing, and claim-level traceability are the main contribution, while the deterministic bidirectional diffusion retriever is positioned as a controlled test case inside that framework. This framing matters for reading the headline tables: the diffusion-and-quota system is not advocated as a new state-of-the-art retriever but as a single, fully specified graph policy that can be measured cleanly against flat dense and graph-aware baselines under the matched interface, so that any observed effect is attributable to traversal rather than to the retriever's identity.