Best-fit models for different choices of pixel size of the ψ_pix component, corresponding to 4, 3, 2, and 1.5 times the data pixel size. The modeling assumptions are otherwise identical to the results presented in Fig. 5 (the second row is exactly the same). From left to right, the columns show: model residuals with reduced χ² value, best-fit ψ_pix model, followed by each scale of the starlet transform of ψ_pix. Each of these scales are images of the same size as ψ_pix, above which we indicate the scale index j and the characteristic scale (in arcsec) of the features. We see that a pixel scale of 3 is sufficient to model the full dynamical range of features supported by the data. Finer pixels are not required, as the smaller starlet scales reveal mainly low-significance artifacts over the level of individual pixels.