The evaluation of the example base starts with setting of assumptions on the behaviour of procedure.
It has to be decided what interval of relevance/tolerance-values counts as being equal to the value in the center of the interval. That is necessary because successively generated tolerance hypothesis differ slightly. The amplitude of the oscillations is used to estimate the progress of the analysis. At the same time, these differences prevent the control analysis from confirming prior analysis by identical results. A precisely defined environment around values makes close values considered equal and enables the procedure to find confirmation in close results of reruns. If the environment is defined too large the procedure finds too early confirmation. By a small environment, the procedure needs many confirmation runs to generate identical results.
As well, a threshold value for the counter values has to be defined. It states how many successive confirmation cycles a tolerance hypothesis requires to be well-founded. The counters vary between the minimum value of 1 and a maximum which has to be defined in advance as well. In between, there must be set a threshold value. Values larger as this threshold indicate a tolerance value that is sufficiently confirmed by combinations of successful examples and tolerance test vectors. The larger the number of required confirmations and the threshold value are, the more reliable is the result of the analysis. A high threshold value comes along with the disadvantage of high computational demand and the risk of no result at all.
Next, there has to be decided what learning behaviour the procedure shows, what impact the result a test cycle on following test cycles has. As the counter values represent the fitness of a tolerance hypothesis, these values have to be adjusted. Prior to entering the core of the procedure, the analyst has to determine the impact of confirming tolerance test cycles on the counter matrix. As well, it has to be decided, how a counter matrix shall be modified after a test cycle with a tolerance hypothesis that can not be confirmed by the current example constellation.
Finally, there needs to be an assumption on the number of intervals between the values of 0 and 1. This determines how precise someone wants to measure tolerance and relevance. This assumption defines the Gestalt of the counter matrix.