Hi Clement,
I have (very) high resolution data, for which direct summation method seems to give significantly better map than FFT (i.e. visible hydrogen vs not).
at some point we spent a lot of time experimenting with this: Acta Cryst. (2004). A60, 19-32 On a fast calculation of structure factors at a subatomic resolution P. V. Afonine and A. Urzhumtsev https://doi.org/10.1107/S0108767303022062 and I can't remember seeing any significant difference in R-factors/maps. So.. there must be something special going on in your case. If you send me files I may be able to tell you what exactly is going on. Otherwise, if you'd like to tackle this yourself, you can start with changing sampling finess (resolution_factor) to something like 1./5 or 1/6., which will still be much faster than direct summation yet produce even greater match between approximations. Pavel