I have realized that the code, as I implemented it in PIPE2D-777 and PIPE2D-778 is wrong. What the code was doing was not summing over the pixel when doing the final downsampling, but just outputing the central value. This was due to me simply copy-pasting the galsim snippet code which correctly applies the offset (interpolating on subpixel values with Lancoz5) and not realizing the difference in this final step. This difference became obvious when using the code provided by Paul in PIPE-872. I have now implemented the proper dowsampling, which is the same as in PIPE-872.

I have to rerun the solutions being prepared for PIPE2D-857, due to this change (see the Figure). Given that the effect on defocused images and Zernike parameters is minimal, I will run shorter runs as we have to significantly modify only few parameters determining the size in focus.