Reconstruction quality of images for our three datasets for different values of λ. The left image shows peak S/N in dB versus λ for the first major cycle for our three initial datasets, illustrating the different ideal values across different objects. This becomes less problematic when normalizing by the L²-norm of the dirty image ||ĩ₁||₂. As shown on the right, $\frac{\lambda }{{\Vert {\tilde{\iota }}_1\Vert }_2}$ is much better behaved, with the region of wellperforming values all roughly lining up. The reconstruction algorithm used here is the single-step sparsity regularized method that reconstructs using all visibilities without partitioning.