Imagine that you are engaged in consulting work for a firm whose goal is to maximize the efficiency of data processing and minimize the complexity of the solutions on which the system is based. The situation is as typical as it is difficult--you must be prepared for hard interviews with the customer, possibly a senior official responsible for decision making, and you must understand the previous motives and experiences of the firm that will sometimes discard your best and most sophisticated ideas. After all, you may soon realize that even your great expertise is not a big thing in solving these numerous problems.

This paper is by two computer scientists who openly recognize that they were initially far from being experts in coding theory (as a result, their reference list consists primarily of rather fundamental titles). They suggest some good recipes for tackling the above paradigm and present the process of consulting for a large mail-order company and elaborating their recommendations. The authors found themselves looking for the most appropriate error detection scheme for a decimal-oriented environment in a system that is supposed to read, and send for further processing, account numbers (of up to several million clients) keyed in by a human. Such input is prone to various decimal mistakes, of which the most common are incorrect digits and the transposition of two adjacent digits.

The authors consider a whole spectrum of different solutions, ranging from the most common approaches, such as the ISBN mod 11 check and the IBM mod 10 check, to more exotic-seeming solutions like mod 97 and mod 997 checks, and assess their capacity to catch randomly generated errors. They give several recommendations, depending on the desired error rate. The authors put some emphasis on an elegant, but overly academic, scheme originally proposed by J. Verhoeff, who also suggested a full list of possible errors [1]. Among the schemes using two check digits for very low error rates, they recommend the mod 97 check with weights successive powers of 10. This check catches 100 percent of each of the errors on Verhoeff’s list.

My overall impression of the presentation is good. It is an excellent instructive example for beginners in the rapidly developing area of scientific consultancy. Finally, one of the key points of the work is “In dealing with commercial firms as a consultant, one should avoid excessively technical or academic recommendations. Companies want workable, simple solutions that they understand and believe in.” To this, I can only add that this recommendation itself should be used in an academically clever way.