Three independently coded risk models produce nearly the same loss estimate. An agency calls the estimate robust because the implementations differ. Which assumption would most seriously damage that robustness claim if negated?