Your problem is likely to be that most docking protocols pass through a centroid phase; Rosetta has no way to represent phosphoserine in the centroid phase and probably converts it to serine. The local_refine method is all fullatom, so the phosphoserine is maintained.
You could try a few things:
A) hack together a centroid parameter file for phosphoserine (um…try just copying serine and changing the name? This is not that valid w/r/t electrostatics or sterics, but centroid mode is pretty vague anyway)
perform centroid docking only, get fullatom outputs of the centroid-docked poses, use a script to patch the phosphate back on, and continue to full atom refinement