They're not wrong answers, they're floating-point answers. What's wrong is the assumption that infinite precision will fit in a handful of bytes.