This is a question about the effect of solvent fraction estimation on the quality of the initial phases.

I am trying to solve the structure of a membrane protein at moderate resolution 3.5A from a MAD data set (peak, inflection and high energies). 

I have one copy of my complex which gives me a solvent fraction of 0.7 without considering the contribution of the detergent.

With a related molecule I found a Phaser-MR solution but it does not refine although it is right. 

I am using Phenix with the phases obtained from molecular replacement and combining them with the MAD signal to locate the seleniums. A visual inspection of the anomalous fourier difference map shows 6 sites out of the 7 expected at 4sigma  contouring.

If I let Phenix decide by itself it goes on its own with  two copies in the ASU and considers a solvent fraction of 0.50; then it finds 7 sites (and not 14 ? why is that?) (the seventh is weak but we expect this for this region of the complex).

now if I force Phenix to take ncs_copies=1 and solvent fraction=0.7 then it only find 6 sites.

Wether it is 2 "freely chosen" molecules/ASU and 0.5 of solvent  or 1 "forced" molecule/ASU and 0.7 of solvent, the 6 first sites are identical (although occupancies are quite higher in the first case)

Should I take into account the detergent and lower somewhat my solvent fraction (in the 0.6 range maybe) with the fixed one copy ASU estimation or just trust Solve/Resolve when it runs the density modification part. Does it really affect the quality of the phasing? Between these two extremes I am surprised to see that the quality of the resolve map are quite similar apparently showing clear new structural features.