Reconstructed potential perturbations for different values of the regularization strength. First row: corresponds to a low regularization strength with λ_ψ, st = 1. Second row: shows our fiducial models of Fig. 5, with λ_ψ, st = 3. Bottom row: corresponds to a high strength with λ_ψ, st = 5. In all cases, the strength for the Battle-Lemarié wavelet (λ_ψ, bl) is set 1σ higher than for the starlet (λ_ψ, st) to penalize isolated spurious features in the reconstructed. The main features of the perturbations are recovered in all panels, so the model is robust against reasonable changes in regularization strengths. A too low regularization strength leads to over-fitting () whereas a higher regularization strength leads to a smoother model, as expected.