So I was drawing probability radii for advancing hurricane positions. I tested my interpolation and extrapolation routines thoroughly, for when NOAA doesn’t provide values they want you to use for specific forecast hours. But in case things go wrong, I picked a small nonzero radius value (0.5) so that if “this should never happen” happens, people who look at maps would see an exact radius that was way too small and come complaining right away. The function does an exact->inexact on the result of a computation, so things that expect floating point are happy. It struck me as ironic that I picked an exact (in my mind) value of 0.5 that is “inexact,” though by IEEE–754 coding, 0.5 is still exact in binary.

https://en.m.wikipedia.org/wiki/Rackett