Dependability evaluation using a multi-criteria decision analysis procedure

System dependability is a complex notion, represented by multiple attributes, and requiring diverse sources of evidence to demonstrate its achievement. The paper argues that the direct measurement of such a composite property is unlikely to be feasible, even for a single stakeholder's view of a system deployed in a specified environment. Instead we propose an indirect evaluation procedure that is based on multiple-criteria decision analysis. The procedure is rigorous and systematic, and we illustrate it through a case study that compares alternative designs for computer assisted braking systems, to find out which of them is the more dependable. The result of the analysis is a theoretically sound, justifiable decision. The approach formalises traditional "engineering judgement" by combining available scientific data with domain expertise, and making explicit the trade-offs within the decision.