The procedure shall quickly adjust the counter matrix to the example base. Consequently, it makes sense not to adjust the counter values linear. Instead, after repeated confirmations less and less is added and after repeated disconfirmations less and less is subtracted. By initially fast and later slow adaptions the counter values oscillate with an decreasing amplitude.
If a tolerance option is represented by a counter value slightly above 1, the FACTORFINDER-Procedure adds a large amount to the counter value after confirmations. High counter values only receive little further increase. The increase decreases until the maximum value of the counter value is reached. At a high level the counter value remains even though further confirmations would suggest a further increase.
Conversely, if a tolerance option is disconfirmed: High counter values are reduced to a large amount. Small counter values receive only small further reduction.
The upper limit on the counter values restricts the effect of perpetual self-confirmations of once confirmed tolerance test vectors. A too large proportion of one counter value in the sum of all counter values of a factor is prevented. Because of that, little confirmed tolerance options get the chance to be selected in future tolerance test vectors.
Confirmed tolerance options run in danger of being repeatedly confirmed. This could stabilize counter distributions within the counter matrix that are far from adequate.
The direct way from such an inadequate representation is not always possible. The procedure has to pass several hurdles: First, tolerance options with little counter values have got a little probability of being selected as tolerance hypothesis. If such a tolerance options gets selected despite this resistance, the next hurdle follows: The new, previously not tested tolerance hypothesis does not fit to the preveiling tolerance hypothesis of all the other dimensions. A factor, that previously has been categorized as being tolerable and unimportant, has generated a specific evaluation of all the other factors. A completely different tolerance hypothesis in one dimension will contradict the preveiling categorization of the other factors. Different tolerance options will only be confirmed together with different tolerance options in other factors.
The direct way from such a situation only rarely is possible. Instead, such a distribution of counters is left by repeated cycles of disconfirmation and a loss of cumulated information. Consequently, the probability of selecting the formerly dominating tolerance options sinks. The chance of further confirmation of these tolerance options is reduced and so the probability of those different tolerance options to be selected increases.