It is a bit tricky but it is the same reason that all meshes has all different reference pose. Although retarget works based on skeleton w.r.t. hierarchy, I couldn’t use skeleton ref pose(which technically isn’t used anywhere else visually). To make it work for retargeting the pose has to be very accurate otherwise it doesn’t retarget correctly, so it uses mesh ref pose, thus the retarget base pose is saved to mesh.
Once we have retarget system work in in run-time, we might not even need skeleton anymore, but we’ll see. That’s a long term plan. 
–Lina,