Wow, that's pretty cool. Translating (almost) arbitrary floating point programs into weird integer programs while also preserving equivalence under non-strict floating point semantics? Mathematics can be surprisingly wonderful.