In the scene vs on the scene is probably the best reasonable definition anybody can come up with for what should be repaired or not. I have no sentimental attachment to dirt or scratches, regardless of what stage of the production they were introduced, so I agree things like that can be safely excised without any harm to the aesthetic of the film, as long as real detail is used to repair the damage rather than the temporal blurring of DNR.
My audio engineering mentor has a philosophy about mistakes in a production, which I think applies to this kind of thing too. There are three options—you can either: 1) get rid of it, 2) ignore it, or 3) embrace it. A truck driving by got in the microphone during the take with the singer's best performance? High-pass filter out the rumble. The filter is making the singer's voice sound too thin? Dial it back to only get rid of the lowest frequencies. Now you can hear the midrange of the truck but not the bass, and it just sounds plain weird? Try some different EQ to see if you can bring it down a bit. Still can't do it without destroying the singer's voice? Turn off the filters and just get used to it being there. Nobody else can hear it but you? Maybe it wasn't much of a problem after all. The truck noise is getting into the reverb send on the vocals and making this neat effect in a key part of the song that wouldn't have been there otherwise? Go with it and maybe even find a way to emphasize it a little more to make it sound even cooler.
Sometimes the little idiosyncrasies of an artistic work can add up to create a whole that was unintended but awesome all the same. Even the mistakes can turn out to contribute to the overall effect. You have to decide whether something like that is enough of a problem to warrant fixing, and also consider whether the result is truly better for being fixed or if the solution does more harm than the original problem. Maybe there was a worse flaw hiding underneath, and by re-compositing or painting over something you've exposed the problem that was previously covered up. Putting a microscope on any work is going to find all kinds of things that aren't perfect, and it's easy to go down rabbit holes trying to fix them. Determine what makes the vibe of the work, and do your utmost to preserve that without getting in the way of things that are important. Don't put yourself above the work in importance—avoid getting caught up in 'improving' things, but figure out how to present it in its best possible light. That's the best anybody can do, and with the proper dedication and tools it should turn out just fine.