The Conclusion explicitly frames the headline tied-at-R@10 finding between adaptive and fixed-depth hierarchical retrieval as a non-rejection of equivalence under the prevailing statistical power bound, not a positive claim that the two policies are equivalent. The power bound is set by the strictest leakage controls, which expose only 21 unique held-out targets on LectureBank-Full and 18 on MOOC-CS across two question families. With this few unique targets, paired-bootstrap intervals are wide and a true small advantage for either policy could not be detected; the correct reading is therefore that the data do not reject equivalence at R@10, while leaving open both directions of a small effect.