Dear authors, congrats on the acceptance-- this paper was discussed extensively, and on balance was accepted. The reviewers provided multiple comments and it will be of critical importance that you revise the manuscript accordingly. One point that the reviewers felt was not deal with satisfactorily is the use of a Gaussian noise model for spike amplitudes, and I agree with that assessment-- at the very least this limitation should be transparently discussed, and work on how to use richer, non-Gaussian prior distributions should at least be cited. Finally, amortized inference and difference distributions on spike localizations/shapes have been studied in the related question of detecting action potentials from various imaging modalities (e.g. by Turaga and colleagues), and I do think it would be useful to link to that literature and the various methods that have been proposed in it.