Chris, I’ve also been careful to say “most”, with you being the unnamed exception .
Later-No-Harm (LNH) is just one of dozens of formal properties a given voting system may or may not satisfy. As you noted, it’s impossible for any system that statisfies the Condorcet criterion to satisfy LNH too, or vice versa. I think it’s fair to say that most people find the Condorcet criterion (“if some candidate beats all other candidates one-on-one, they’re the winner”) far more compelling.
Indeed, IRV is one of the only systems in actual use that satisfies LNH. Perhaps the only? Random Ballot satisfies LNH too, but that’s not exactly a good argument for it
And it’s not true that adding preferences to an IRV ballot can’t “hurt” you in other ways. You can’t hurt your favorite, but you can hurt your lesser preferences’ chances of winning.
To me, the essence of your complaint is that if you’re truthful about your preferences in other systems, that may cause your favorite to lose under other systems. The fundamental problem is that I see that as a good thing: if, overall, society prefers some other candidate to my favorite, they probably should win. If, e.g., I slightly prefer A to B and am truthful about that, and B wins “because” I was truthful about that, no, they didn’t: B won because the voters as a whole, including but not limited to me, preferred B overall.
IRV intentionally blinds itself to the totality of information voters give, dribbling it in to the process one round at a time. This allows it to satisfy LNH (your lower preferences are invisible to it until your favorite is eliminated), but at quite a cost.