Stacked residuals between our non-Gaussian (fit) models and the mocks. Fitting the non-Gaussian model using the |$\mathcal {L}_{1}$| likelihood significantly reduces the stacked residuals relative to the (unfit) Gaussian model (see Fig. 5), while fitting with the |$\mathcal {L}_{2}$| likelihood effectively eliminates the residuals. The absence of residuals indicates that the model precision matrix could replace the mock precision matrix in a variety of applications.