Model residuals from the Adaptive+Matérn model and central region of the true source (top row, as in Fig. 2). Suppressing the flux in the central most bright pixels (> 0.3 in Fig. 5) by 50 and 95% (middle and bottom rows), creating new mocks, and modeling them with the same setup, leads to improved residuals. The plain semi-linear inversion technique using just a regularization of the form given in Eq. (10) cannot adequately capture point-like, cuspy features in the source light profile. More advanced schemes, like a luminosity-weighted regularization scheme (see Sect. 3.3), perform better in this case.