The Conclusion explicitly scopes the paper's contribution as methodological rather than algorithmic: deterministic bidirectional diffusion with role-aware quotas is presented as a controlled test case for auditing strict-parity evaluation of prerequisite-graph retrieval, not as a proposal for a broadly superior new retriever. The graph-policy variants are vehicles for stress-testing the evaluation contract (strict parity, leakage audits, separate token-cap diagnostics, external-validity boundary checks, traceable artifacts), and the headline retrieval numbers should be read as evidence about what those controls reveal rather than as a recommendation to deploy diffusion-based retrieval in general settings.