draws need to be resolved
- measure in cluster consensus e.g. as in-clustervariance (will not address the case of pass/fail concepts
- in case there are two clusters of the same size and same inClusterVariance, choose the one with the lower score (fail!)