In practice, you get guarantees “up to precision”. This is totally reasonable. If you’re doubtful, you can either prove theorems, or bump up the precision.

In general this works well.