In a synthetic data experiment, we generated a spike train for a population of Ctrue = 50 cells and Ttrue = 2000 time bins. Then we varied the number of neurons C, recording duration T, scale of the firing rate λ, temporal bin size Δt, and inferred the number of inferred latent states from the data. Horizontal dashed lines indicate the ground truth. Box plots are obtained from 10 independent Monte Carlo chains, and each chain was run for 1000 iterations; the number of states in the last iteration is used.