t-SNE plots comparing the observed velocity maps of four galaxy clusters to the TNG simulation library. While all embeddings were generated using a single trained Siamese CNN, we present independent t-SNE projections for each source to clearly visualize the specific matches within the broader simulated distribution. Points represent the relative similarity of velocity maps in the latent space, with proximity indicating a closer kinematic match. The nonuniform distribution of the TNG-simulated clusters reflects underlying physical gradients and evolutionary sequences learned by the network. The axes are arbitrary, and quantitative similarity rankings (blue) are calculated based on Euclidean distances in the original high-dimensional embedding space rather than these 2D stochastic projections.