Recently I have seen a paper,"software project and quality modelling using Bayesian networks". In the paper they divide software process models into two broad categories: cost models and quality models. My question: during modeling software testing process, when to choose metrics for collecting data, may I combine cost model and quality model, or only consider one of them? And another question, is it a must step to consider what kind of metrics to guide the collecting data? Can I generate metrics from the real process?