Hi Anne,
Thanks for the quick reply. I downloaded the files, and I got the result you predicted - all the occupancies looked neat and tidy as expected. Since you used those nice occupancies to generate perfect data, which was then refined against, that's what one would expect, right?
The occupancies in starting model, "start.pdb", are all garbage. So, if phenix.refine does something wrong with occupancy refinement, they would never refine to correct values.
So I am still left with the issue of the occupancies going nuts in my real refinement.
The test I did completely replicate the real situation: obviously, phenix.refine "does not know" that data.mtz contained fake Fobs -:) If you are still getting the wrong values then there must be something else going not right. The only way to find out is if you send me the data and model and the exact command you used. In this case I will be able to reproduce the problem myself, investigate some more and hopefully provide a fix. If you decide to send me the data, please make sure you send it to my address only and not to the whole bb. Pavel.