A model is proposed to predict the integrity of work done in a software development process i.e. the model monitors the quality of intermediate products during a development process, to guide technical project management decisions. In particular, an aim of the model is to identify degrading quality levels as early as possible, so that reviews can be instigated to correct the problems before their consequences 'fan-out.' The model decomposes the development process into a skeleton of sub- developments. For each sub-development, the input to the model is a set of measurements of the products and processes involved. The method is based on graphical probability models.