In a sense, subtraction and division don’t really exist as full fledged operations on the reals anyway. They’re just convenient notation for additive/multiplicative inverses.

The reals are irrelevant to a machine, because Almost All Reals Are Normal you definitely can't do real arithmetic with a machine.

Floating point is just the integers wearing a funny hat. Fixed point is a different funny hat. The rationals are two integers in a trenchcoat.

You can do a meaningful arithmetic with any real you're able to enter into the machine [0], and the tidbit they took the time to share makes sense even in that sort of restricted context.

[0] https://blog.acolyer.org/2020/10/02/toward-an-api-for-the-re...

That's actually a really interesting piece of software, thanks for the link!