ordering operations poorly so errors build up | - if a numerical application is designed to work with floating-point numbers, then make sure it works with inputs that vary widely in magnitude, including both large positive and large negative exponents
- pay particular attention to the accuracy of the result when a floating point value is being repeatedly decremented or incremented by small amounts
| when you do small operations on large floating point numbers, and excessive rounding or truncation errors build up | defect in a numerical algorithm |