- Current implementation (clusterCrit) is fast on it's own but does not reuse distance matrices that could be determined only once. - Only euclidean metric is supported, extension to support arbitrary metrics would make the internal cluster indices more meaningful.