The achievement of accurate software cost estimation based only on a few factors is a long-standing goal in software engineering. Work in this area is exemplified by a number of algorithmic approaches that have been proposed over the years. COCOMO is one of the most frequently quoted of such approaches. Evidence emerging from observational and simulation studies suggest that feedback mechanisms play an important role in determining software process behaviour, its dynamics and performance. Thus, the presence of feedback mechanisms, and in particular feedback control may have a significant influence on software project cost and interval performance, but none of the current algorithmic cost estimation approaches appears, at least explicitly, to account for such influence. Why, in spite of this, do algorithmic approaches provide satisfactory estimates? Why did they work? This paper discusses some possible answers, that at the present must only be taken as hypotheses. The paper provides suggestions for further investigation of the problem.