Sure we add up the numbers, but there are thresholds for everything anyway. We might not be concerned about something within a 1MM range let alone a floating point inaccuracy. The uncertainty is accounted for already.