Foreground bias and custom analysis mask related to needlet-based component separation. Left panel: Foreground residuals in the reconstructed CMB power spectra in the inhomogeneous goal-pessimistic noise scenario, with a fiducial input cosmology of r = 0, A_lens = 1, and Gaussian and d0s0 foregrounds. We show the reconstructed CMB and foreground power spectrum residuals $C_{\ell }^{\mathit{BB}}$. The residual is calculated by mixing the pure foreground maps with the NILC weights. Right panel: constrained analysis mask used by pipeline B in the case of Gaussian foregrounds, built from the fiducial mask shown in Fig. 2.