If your experimental phases are weak, your model bias will be strong
This is the duck
Those are the x-ray
data from duck
This is the cat
(which you thought
is the right model)
This is duck data
phased with cat
As bad phases (wrong model) overpower the experimental data, you
want initial phases with a high figure of merit (probability of
correctness) to as high a resolution as possible.
The more model you can build into the initial experimental map (the
only PURE map you ever have) the less model bias you will introduce
upon phase combination.
Donít have a cat.
Acknowledgement : The Fourier duck idea is borrowed from Kevin Cowtan
Biomolecular Crystallography - Overview