Modeling the "good enough to release" decision using V&V preference structures and Bayesian belief networks

Throughout the process of determining when a unique computer-based system is "good enough to release," an assessor must consider and reconcile process and product evidence as well as make a judgment on the severity of remaining faults. The assessor may be working with uncertain or incomplete knowledge, and may have little data by which the evidence can be validated and verified. As well, the assessment may be done on an ad-hoc basis with unstated or untested assumptions concerning the relative importance of evidence. It can be difficult to repeat ad-hoc or loosely structured assessments with any degree of confidence, and it may be near impossible to recreate a given assessment for an audit. A model of the "good enough to release" decision based upon quasi-order preference structures of validation and verification (V&V) activities is proposed in this paper. We focus on modeling the release decision for unique computer-based systems because of the types of evidence assessed during the decision. We use quasi-order preference structures to determine the V&V activities that are generally considered to be the most effective, and to determine relationships among the activities. We use a Bayesian Belief Network (BBN) as the modeling formalism because a BBN's characteristics support the type of assessment process being modeled.