Formalising a software safety case via belief networks

Belief Networks (also known as Graphical Probabilistic Networks and with various other names) offer a useful fonnallanguage for stating complex arguments in rigorous, yet visually clear terms. They are thus promising candidates for describing the complex, often unclear reasoning that is often implied, but not described, when reasoning about software dependability, in particular when "engineering judgement" comes into play. We introduce the problem of building a rigorous safety case for software, and argue the merits of belief networks as an aid for building, criticising and perfecting such safety cases. This first report includes a high-level introduction to Belief Networks, and then introduces and discusses a small but realistic example. Our conclusion is that this method has great potential for making safety arguments easier to communicate and check, and in the end more trustworthy.