One of the things that confuses the issue even more is the “fortuitous discovery”, namely, any of the players/singers/engineers will occasionally do something that was not planned, but definitely seems to improve the song, such as a little tasty on the guitar or drums, a fader or panning change at precisely a great but initially unplanned time, etc. Our repeated listening solidifies these things as part of the song in our minds. If we should later re-record the song and one or more of these things we liked are not there, the song sounds diminished, at least in our mind. The difficulty is being objective enough to truly evaluate the overall re-recorded song rather than just a few things newly added or subtracted from it. But that’s not easy if we’re missing something we really liked.
My approach to improving a song has often been to remove the things I don’t like (a mistake, a dissonant note or chord I or someone else thought would work, etc.) That’s especially easy with something like Melodyne, where I’ve even changed a chord from minor to major or in reverse as well as making a chord a 7th, 9th, etc. after the fact. HOWEVER, if the song was a turd, it’s still a turd, even though repeated listening got us used to it, perhaps even in love with it. It’s hard to be objective in those situations, because to everyone who hasn’t been listening to this song over and over, it’s still a turd. And as the saying goes, you really can’t polish one.
MJ’s comment that he can re-listen to a song objectively even though he’s heard it many times before is truly a powerful ability – I don’t think I have that ability, though I’ve never truly tested it and this discussion tells me that I really should try. If I can do it, I’ll have a achieved something valuable. If I should instead learn that I can’t do it, then at least I know that I need the ears of others to avoid creating the aforementioned occasional “turd”.