Most of these simulations were carried out using CLEAN to deconvolve data corrupted by a single realization of the noise. How robust are conclusions based on the measured qualities of these images to different noise realizations, and different deconvolution algorithms?