I don't think there's any in-depth documentation on how scaling works per se, but just from PQ code it simply scales the character's transform by the value calculated from the position in the current region, and then the pixel-perfect camera effectively applies the "nearest neighbour" interpolation to the result. So the unpredictable result at lower scales is sorta expected.
Don't think there are built-in solutions for your case. And in case of overlapping scale regions, it looks like instead of combining the scales, it uses the values from one of them, which is not really useful.
I'd recommend joining the Discord server and asking in the general chat or creating a topic, the discussion over there is much more active