I'm sure the answer is "it depends on your experience level", but is there a general rule-of-of thumb on how thin is too thin?
Well, everyone has their system and opinion, and of course also everything heavily depends on the actual circumstances, but generally what I'd do is:
1. Measure the thickness of the paint at the door sills around the car.
2. Take the highest value I measured there. If it's below 60 microns (~2.3 mils), I'd take 60 microns instead.
3. Multiply the thickness I got in step 2 with 1.5x.
The resulting paint thickness would be about 2/3 of the thickness of paint at fully clear coated surface, and the thickness I'd never want to go below with any heavily abrasive process, like wet sanding and compounding. (Polishing is a different story, because that only takes 1-2 or two microns at most, so you don't have to watch out that much when doing that. That said, I wouldn't try to polish anything under 80 microns either, unless I - or the person I'm polishing the car for - would be willing to risk compromising the paint to a degree where the car might need a repaint.
Of course this is only mostly true for cars which had an original (untouched) factory clear coat thickness of at least >110-120 microns or more. Some of the newest cars might have a total paint thickness of 100 microns or even less right out of the factory, which of course leaves you with even less wiggle room.
If I'd know that the scratch or scratches I'm trying to remove are likely deeper than what's left of my wiggle room, then I wouldn't even try to go up to (well, down to) the limit calculated above, because it just makes no sense to abrade all the remaining "safely" removable clear coat, when in the end you won't be able to take out the scratch anyway. Then the most you can do is just diminishing it, most of which you can do in the first possibly 10 microns or less. So, I'd only go that far there, even if my calculation would allow for more cc getting abraded.
Also note that if you can measure heavy variances in the paint thickness on a panel (like >5-8% variation) then that panel either has been repainted, was somehow defect already from factory, or has been already extensively spot-corrected, which all would increase the chance of you running over an area that has been already either partially compromised, or where the actual paint thickness could be far less than elsewhere on the panel, even if measurements show it to be pretty high a that spot - because there might be also body fillers or uneven panel surface below the paint. Such a panel would be even more risky to touch, so I'd would possibly add another at least 10 microns to my minimum limit I wouldn't want to go below on that particular panel.
Even then there'd be a slight risk that you're burning through the clear coat or compromising it, just like there always is, under any circumstances, no matter how thick the paint seems to be. So, you always have to "count" with that, regardless of not going with that assumption (ie. that you will burn through at an unexpectedly high remaining thickness already).
