Acceptance criteria for critical software based on testability estimates and test results

Testability is defined as the probability that a program will fail a test, conditional on the program containing some fault. In this paper, we show that statements about the testability of a program can be more simply described in terms of assumptions on the probability distribution of the failure intensity of the program. We can thus state general acceptance conditions in clear mathematical terms using Bayesian inference. We develop two scenarios, one for software for which the reliability requirements are that the software must be completely fault-free, and another for requirements stated as an upper bound on the acceptable failure probability.