Thanks for a masterfully written article, with an excellent overview of the problem and the language design choice.
It makes sense to have to opt-in to floats for performance in domains like scientific computing or graphics (where silent precision loss matters less). But in all other cases business logic (money etc.) would give you normal arithmetic as you would expect.
It also makes sense to not allow numbers to overflow to the Heap, but rather have the programmer take special care for exceptionally high numbers.
Having no shenanigans between the compiler output and the runtime output is also a sane choice (staying true to predictability / determinism).
All in all, it sounds like good language design choices for a Web Server and Business Applications focused language such as Roc.
