The FACTORFINDER-Software needs a record of all acceptable examples as a data base. This example base shall be as complete as feasible. Several successful variants must be known so that comparisons between acceptable variants are possible. Inferences leading to relevance values are essentially based on the assumption, that at least one example shows each ideal value and therefore the ideal values are known to the analyst.
If a factor shows an ideal value and is less than critically important, then there is at least one other variant recorded in the data base that differs in this factor and shows a non-ideal value. That follows from the assumption that from each ideal value at least one example is known and entered into the example base. It would be simple if it would be known in advance what values are ideal and what not. But this is not known with certainty. The only fact that is known to the analyst is that all non-ideal values of an acceptable variant are not important enough to make that variant fail.
The FACTORFINDER-Procedure compaires each acceptable alternative with all the remaining acceptable variants pair-by-pair. From these comparisons hypothesis concerning the relevance of the factors are derived. The procedure starts with a randomly generated hypothesis concerning the tolerability of non-ideal values in every factor. This hypothesis is checked against the available data. A recursive procedure of preliminary interpretations of data, evaluations of this interpretation and new re-interpretations follows.
This procedure generates an evolution of perpetually improved tolerance hypothesis. A matrix of counter values serves as a memory and buffer for competing tolerance hypothesis. All plausible and partially contradicting hypothesis on the tolerability of non-ideal values are melting in this counter matrix. The counter matrix in itself represents the tolerability of non-ideal values in each factor. It restricts the consequences of the number of possible combinations of tolerance values. It prefers formerly successful hypothesis when selecting new hypothesis and infuses random new hypothesis to a certain extend. Information on the relevance of the factors and former confirmations of this information are saved in the counter matrix.
As a result of the FACTORFINDER-Procedure, a classification of all alternatives is generated. With a sufficiently large example base, the procedure identifies those variants that show ideal values. The procedure fails when searching for a relevance hypothesis as long as the available example base is too heterogeneous. This property is used to identify those examples that strongly differ from all other variants.
The FACTORFINDER-Procedure assigns relevance values to each factor. These values help to evaluate alternatives, to compare variants and to judge the success of alternatives that have not been tested in reality yet.