When Model-based Testing Fails

Armando Haeberer made a substantial contribution in translating results from the philosophy of science to software testing. He showed that the limits of the standard methodology of confirming theories, where observations are derived from the theory and hypothesis, are also present in software testing. He criticized the classical test case generation approach from algebraic specifications, but also supplied an alternative based on Glymour's bootstrap approach. However, the debate that one would expect after his contribution never really happened. This is especially astonishing since most of the model-based testing approaches follow the method he criticized. This paper aims to open the debate by looking on his findings from a fresh point of view. We use the concept of refinement to clarify the fundamental issues involved in testing and discuss the practical consequences.