The Convergence of ‘Similarities’ and Making the Best of Probabilistic Evidence