like listen, maybe this doesn't even belong in euclidean space, you know? we should probably be treating this as a sequence/time-dependent model, where the selection of each name at each step covaries with the probability of selection at all previous steps