Incorrect comparison of dyadic constants or something like that

[bmsherman]

# 0.1 - 0.1 < 1e-21;;
- : prop = False

This is obviously incorrect. I'm not sure what's going wrong.

[bmsherman]

I think it could be that my decision to represent uncertain data (i.e., not-exactly-representable constants) as intervals doesn't play correctly with A.upper, where I probably actually need to reverse the interval in that case. I'll need to think about it.