It sounds very reasonable that edits made in ACR before conversion are non-destructive
But I still don't get the point. What do you mean by saying "edits made un ACR before conversion"? with "conversion" you mean to perform the whole RAW developing process and import the resulting 16-bit TIFF into PS?
What's the difference between playing with ACR's curves until you get the desired result, and playing with a mask layer in PS until you get the desired result? is the first non-destructive and the second destructive? If so I disagree with this.
If ACR shows you the effect of let's say a curve, is because has performed the whole RAW developing process at least on the area currently displayed. And that means Bayer interpolation, and pixel levels alteration. Otherwhise it simply couldn't be displayed.
ACR is not going to perform all its adjustments at a time over the RAW file. They look like that to the user, but internally what ACR does is a completely sequential process where, for speed and simlpicity reasons, adjustments are applied one after another. White balance and exposure control are first, demosaicing comes next, hightlight recovery (if activated) is next, then comes colour profile conversion, curve adjustments,...
Some of the elements can be applied in differente positions obtaining advantages in some cases, but finally the process is a sequence.
The same happens in PS when many mask layers are used. I find it difficult to believe that PS summarizes all of them before applying them to the image, avoiding error propagation for rounding. Layer adjustments are applied one after another; but still 16 bit (actually 15) is still good enough to keep a high level of quality in the results.