At the beginning of each test cycle the memory that later will save the result of the test is cleared. Then, the candidate example together with the tolerance test vector are compared with all reference examples. This test reveils whether at least one other example confirmes the test vector. For this test, a first reference example is selected. It must be an acceptable variant, because unsuccessful examples only give a weak confirmation for tolerance hypothesis. The focus of the forthcoming evaluation is now on those factors that differ between the candidate and reference example.
The tolerance hypothesis of all factors that differ between candidate and reference example are added. If the sum exceeds 1.0, the current tolerance test vector appears to be inadequate. The procedure switches over to the next successful reference example. If the sum of all tolerance hypothesis of all factors that differ stays below 1.0, this reference example confirms the current tolerance test vector in those factors that differ. Otherwise, if the sum exceeds the value of 1.0, the procedure switches again to the next example.
Very rarely, the first reference example confirms the tolerance test vector in all factors. This would require, that candidate example and reference example differ in all dimensions and the sum of all tolerance values is below 1.0. Normally, it is recorded, what element of the tolerance test vector is confirmed by each example. Afterwards, the procedure turns to the next reference example until all successful alternatives are compared. As a result of this test cycle, a list of all confirmed dimensions of the tolerance test vector is created.
The test vector is tested by comparing candidate example and reference examples. The result of this evaluation serves to improve the counter matrix. After many adaptions for each candidate a counter matrix forms that reflects the relationship between candidate example, reference examples and target criterion.
The tests of the tolerance hypothesis are based on those factors that differ between candidate and reference example, not on those factors that coincide. Using coinciding factors would be helpful if all dimensions would be of little relevance. In this case, only few ideal values would be enough to turn examples successful. Many variants would be acceptable and surpass the threshold of the target criterion although their values are predominantly non-ideal. As long as factors are analysed that are expected to be important, it is more promising to base inferences on factors that differ between candidate and reference example.
The comparisons between candidate and reference example lead to inferences on the plausibility of the tolerance test vector. A tolerance hypothesis is plausible, if there is a reference example belonging to a candidate example, that is successful as the candidate and the sum of the tolerance values of the factors that differ is below 1.0.
This can be shortly explained by the assumption, that the list of successful examples is supposed to be complete. Without this assumptions, the procedure could not infer anything from the example base. If no acceptable example is missing in the example base, then there is always a second other successful example that differs subcritically. In important dimensions, only few reference examples differ subcritically from the candidate example. This is the consequence as one of the examples (either the candidate or the reference) must consist of a non-ideal value. At the same time, if an important factor shows a non-ideal value, only very few other dimensions of that pair of examples differ.