The most common failure is testing round, clean amounts and missing the values that land right on a rounding boundary, such as one ending in .005 after a discount or tax calculation. These boundary values expose differences between rounding rules, and between the mathematically correct result and what a naive floating-point implementation returns.