The paper's error analysis identifies seed quality (how well the dense encoder and query interface surface the right initial candidates) and graph noise (missing, spurious, or misaligned prerequisite edges, including bilingual aliasing) as the main unresolved bottlenecks for hierarchical prerequisite retrieval. Depth control — including adaptive depth gating — is not the limiting factor: on LectureBank-Full, residual misses are near-misses along the local prerequisite graph, leaving little headroom for a depth-control fix; on MOOC-CS, residual failures are dominated by distant misses and bilingual aliasing, neither of which is fixed by varying traversal depth.