The sibling essentially sees what was going on here, I was thinking of a closely-weighted linear classfier where a small change to the weights could alter their ordering but do little to change the results.
Although it's definitely my responsibility to write clearly, in the future I'll think more carefully about what my replies are actually disagreeing with before I make edits that I think are just phrasing!
(And yes, I also had a few edits here before I found a way to write it clearly.)