The paper's abstract states that its contribution is methodological rather than algorithmic. The main novelty is named as three pillars: (i) strict-parity graph-RAG evaluation, (ii) leakage auditing, and (iii) claim-to-artifact traceability. Deterministic bidirectional diffusion with role-aware quotas is positioned as a controlled test case, and adaptive depth gating as an optional extension, not as headline algorithmic contributions. This framing constrains how the paper's empirical results should be read: gains and ties are reported to support methodological lessons about how graph-RAG comparisons should be conducted, not to advocate a new retriever.