The simple explanation (I hope) is that by ranking a candidate lower, you can promote a second candidate so that they survive elimination instead eliminating a third candidate, which would otherwise have defeated your first candidate in a later round where the second candidate will not.

It is a consequence of the loser-elimination step. You can avoid it by eliminating loser elimination, which gives the (unlimited ranks variant of) Bucklin (with ties and truncated ballots allowed, essentially Majority Judgement) voting system, which has its own pros and cons. I kind of lean towards unlimited ranks, ties-allowed Bucklin—count first-place votes, if you have a candidate with a majority of ballots cast, they win, if not add all candidates second place votes and see if you have a majority (of ballots, not counted votes which will now exceed ballots), etc., with a tie-breaking procedure if you get two majorities in the same round (bigger majority, either in current or subsequent rounds, random selection of the tie persists when ballot preferences are exhausted)—is my preferred simple ranked-choice system. Condorcet methods arguably choose a better winner in the case where they disagree, but are less simple.