The author introduces the notion of convergence in a rule base as an alternative to completeness. He begins with a brief review of some examples of rule collections for programming. The examples are taken from various application areas. They illustrate his point that completeness is not a useful notion as a measure of utility in knowledge-based program development.

The question he addresses in this paper, however, is: Does the process by which programming rule sets are developed converge in some sense? The example employed to demonstrate the convergence notion was a rule base of 400 rules for symbolic programming and a control structure for applying them to the implementation of abstract algorithms. Each rule is able to refine an algorithmic statement only slightly, but successive application eventually yields a concrete LISP implementation of that algorithm. An examination of the pragmatic development of this rule base (called PECOS) against its application history shows a saturation curve for the number of rules required to complete a program versus the number of individual programs developed.

While this paper is admittedly subjective, it is very insightful and has all the earmarks of being one of those seminal papers that will often be referred to in the future. The tentative conclusion offered by the author (who is already very experienced in the field of knowledge-based program construction) is that a process of developing programming rules will never cease, but will converge to a generally recognized set of rules that have high utility over different application domains.