Hi Mischa,
The comment was prompted by their Refmac R-factor test failing.
I guess this is the root of the problem, in the first place. It is strange to test a file content with a refinement program. When I was depositing a structure a while ago it also failed that test, obviously because that program does not have a neutron scattering dictionary. For PDB deposition you can use any of two files created by phenix.refine: xxx_data.mtz or the one that has Fmodel and map coefficients. Both files contain the original reflection data (Iobs or Fobs - whatever the input was) and Rfree flags. It is very unlikely that there is a bug in any of these files since they are so heavily used in so many places for so long time... Also, you can use phenix.mtz.dump file.mtz to see the content of file.mtz file. Anyway, if you suspect a problem, please send me the file and I will quickly investigate. Pavel