> That’s it’s by far the best fixed-size approximation to real number arithmetic we have (*)

* for some use cases (**)

** conditions apply