Supplementary error-taxonomy counts for MOOC-CS show that residual retrieval failures are dominated by distant misses — the retrieved concept is far from the gold target in the prerequisite graph — and by bilingual aliasing, where Chinese and English surface forms of the same concept are not aligned. Both failure modes are upstream of any traversal-depth choice, so they implicate seed quality and graph noise (e.g., the English-template/Chinese-concept-name mismatch and noisy edges) rather than depth control as the remaining MOOC-CS bottlenecks.