Best fit χ²/v as a function of the ratio between the correlation length scale (proportional to 1/ℓ) and the period, T. The χ² was computed for fits of Eq. (4) to data perturbed by draws from the covariance matrix, varying the correlation length scale. For each ℓ, 25 Multinest retrievals were run in order to compute the uncertainty on the χ², shown as the shaded region around the mean. In blue, the covariance is properly accounted for in the likelihood, while in orange only the diagonal of the covariance is used in the likelihood. In order for the reduced χ² to be a useful metric, the covariance must be properly accounted for.