Accurate Estimates without Calibration?

Most process models calibrate their internal settings using historical data. Collecting such data collection is an expensive, tedious, and often incomplete process. Hence, we seek an alternative to calibration from historical data. Formally, historical data offers constraints to a set of model options. An alternative methods of generating those constraints is to augment a process model with a search engine (simulated annealing plus variable ranking) that finds options that improve the process model outputs. The improvements can be dramatic: in one of the studies in this paper, the search engine found process options that reduced the median and variance of the effort estimates by a factor of 20. In ten case studies, we show that the estimates generated in this manner can be quite accurate, despite being generated without using historical data.