A working theory can involve a compromise between pragmatic application and pure theory. This is illustrated by the development of Coombs's theory of data. The original version of the theory purported to guide investigators in the use of scaling techniques, but subsequent revisions including a logico-mathematical statement of the theory do not show whether the practical recommendations made earlier still apply. These practical recommendations, some of which Coombs has discarded to obtain a comprehensive theory, are important in spite of shortcomings (...) as they reveal differences between attitude and mental test data. The different forms of Coombs's theory point to the existence of a general problem, how to find a compromise between practice and pure theory. (shrink)