Some issues in multi-phase software reliability modeling

During early software testing phases, testing profiles are often very different from operational profiles. Consequently, assessment of operational software quality during these non-operational testing stages is difficult, and is open to interpretation. The paper discusses some issues related to this. Software is assumed to be a large system composed of components that evolve in parallel. The focus is on early identification of software components that in operation may be excessively error-prone. The approach involves definition of states based on static and dynamic propositions about the verification and testing history of the software, and the use of that information in models that span multiple testing phases. An example based on a risk model is presented.