7D latent space from a NN using the full architecture trained on all BAHAMAS and DARKSKIES simulations, with all simulations treated as known and X-ray maps included. The diagonal subplots show the distribution of values from each simulation in each dimension, while the off-diagonal show the correlation between each dimension pair. The NN was trained with an information-ordered bottleneck, resulting in the last dimensions (right-most columns and bottom-most rows) being biased towards containing the least important information. The first dimension is enforced to correspond to σ_DM/m via Equation 3. The first three dimensions show the majority of the physical information, showing σ_DM/m, the level of AGN feedback, and the separation of BAHAMAS from DARKSKIES, while the later dimensions are modifications of the earlier dimensions, structure dimensions to enable simulations forced apart in other dimensions to be brought closer globally, or Gaussian-like dimensions where little information is encoded.