When you specify a deviation tolerance on linear dimensions, you enter ".002" in the "Upper" tolerance box, and ".001" in the "Lower" tolerance box, and you get X +.002/-.001. When you specify a deviation tolerance on a hole size (specified using the Hole/Thread Note tool), if you enter ".002" and ".001", you get X +.002/+.001 (both values '+'). You need to enter ".002" and "-.001".
Having different rules apply to such similar operations is inherently dangerous. It's my mistake, yes... but it was very easy to make. I understand there are applications for this, I'd just like to see the same method/thinking applied to both operations.