If psychological judgments of similarity are non-metric (Tversky, 1977), and we are evaluating similarity of neural representations by using metric spaces (like Procrustes distance), then how does that mismatch get resolved for neural representations of similarity?