Results from fitting our Markov switching models to our measured N_p time series data from the entire 2023-04-24 single-pulse stack. We show the best fit of our three-state model with first-order autoregression and a globally constant variance. The panels show, from top to bottom, the N_p time series data, the best-fitting noise-free switching model, the model with added Gaussian white noise with the estimated variance (M+N), the residuals of that (data – M+N), the state probability time series p(S​_t), and the most likely emission state for each rotation coloured separately. Red corresponds to S​₀, orange to S​₁, and green to S​₂. We show the best-fitting state means with dashed lines of the same colour in the model panel. As above, the red vertical markers highlight rotations that were RFI excised. The model describes the data well, with almost white residuals.