By the help of the relevance vector those variants are identified where success comes unexpectedly or where failure could not be infered from evaluation of the other examples.
An example base that is incomplete because of missing acceptable variants leaves certain tolerance test vectors unconfirmed. After inclusion of those missing successful examples their confirmation would be possible. Because the examples are missing, the relevance is overestimated.
The relevance is underestimated at factors where each ideal value is complemented by more than one non-ideal value. This factors seems to be more tolerable than justified because non-ideal values in the candidate examples are confirmed by non-ideal values in the reference examples. This impairment of the relevance estimate can be put into perspective: The more important factors are, the less different values can be found in acceptable variants. Therefore, this restriction holds true predominantly for less important factors in later steps of the analysis. Then the example base comes close to a complete picture of all possible examples.
It is quite easy to check whether the relevance is underestimated: If the sum of the relevance values of all factors is smaller than 1.0, all thinkable combinations of values will be expected as a success. This low relevance would make any strive for better alternatives superfluous because any would be acceptable. Therefore, an adequate data base can be expected if the sum of all relevance values of all factors is well above 1.0. Then ideal values appear to be important and non-ideal values are not unlimited tolerable.
Doubts on the generated relevance vector develop as well, if an inacceptable example differs from an ideal alternatives only in dimensions that totally look subcritical. That means dimensions are unterestimated and the relevance is inadequate.