Best-fit models compared to the ground truth spectrum, in order to determine how well the retrieval can account for the systematics introduced through the data analysis. The residuals are calculated by sampling the posterior distributions to generate spectra, and taking the standard deviation at each wavelength. Left: the best fit from the retrievals for each KLIP (blue), PynPoint (orange) and ANDROMEDA (green). Right: the best fits of the KLIP retrievals with (blue) and without (yellow) including the covariance matrix, as well as the retrieval on the Gaussian noise (red) and Noise-free datasets (green). These are again compared to the true input spectrum.