Well, one answer is to use IEEE 1788 interval arithmetic, which will at least give you IEEE 754 high and low bounds for a calculation, rather than one answer th

Well, one answer is to use IEEE 1788 interval arithmetic, which will at least give you IEEE 754 high and low bounds for a calculation, rather than one answer that's clearly wrong. Otherwise, some inaccuracy is the trade-off for fast floating point calculations.