Iteration count is a model choice

Richardson–Lucy is iterative, and more iterations do not automatically mean a better reconstruction. Past some point the method starts to emphasize noise.

The notebook’s explorer opens at 20 iterations (interior error 18.9% → 13.1%), and a second explorer computes the interior error for every count from 1 to 40 iterations and marks the best one tested, so you can see whether the method is still improving or has started to amplify noise.

Richardson–Lucy does not restore the exact hidden truth. It produces an estimate under an image-formation model.