Mismatch between the data and the data modeling. Top: divergence between the data covariance and input parameters. The large mismatch at low ℓ is due to correlations between the input component maps. Bottom: divergence between the data covariance and the recovered parameters. During sampling, we impose no cross-correlation. Thus, the sampler converges towards components that are uncorrelated but whose power spectra are almost capable of capturing the covariance of the input component maps. The red line is the mean of the expected chi-squared distribution that D_ℓ should approximate.