Treffer: Pick a Number--Any Number.

This article details the efforts made by the author to evaluate Epsilon's software development team, analyze deficiencies with the legacy products, and set goals for their replacement. To create a believable estimate, we go through three exercises. The first is a low-effort, low-accuracy quick-and-dirty estimate we call a scientific guess. These typical ambush estimates take the form of a point-blank question on how long this will take. The responses vary from 20 to 120 staff-months. The second estimation exercise is more in the spirit of the wide-band Delphi approach. The developers once again are asked to produce an estimate, only to spend more time and effort on it, going through the estimation process as if responding to a customer change request. They create a statement of work and then estimate the effort. The range is 75 to 150 staff-months. The major drawback of these dead reckoning style estimates is that they are still point estimates--which really do not represent a probability distribution. In April 2005, we have conducted a small experiment. One engineer has worked with us to develop the techniques and rules for applying the classic Java and C++ refactoring rules to C. Thus, we have determined that for Epsilon's system, an average programmer could refractor at a rate of about 250 source lines of code per day.