Quality of the X-ray Data

The data normally should be of high resolution. Unrestrained xyzB refinement with ARP/wARP at lower resolution can potentially lead to a poorer quality density map. The X-ray data should be complete, especially in the low resolution range (5 Å and lower). Ideally the X-ray data should have no low resolution cutoff. Typical symptoms of incorrectly scaled low angle data when running ARP/wARP are slow convergence, higher R factor and R $_{\rm free}$ and building an incomplete protein model (warpNtrace) or too few solvent sites arp_solvent. If the low resolution strong data are systematically incomplete (e.g. missing or overloaded reflections), the density map, even in the case of a good model, is usually discontinuous and is inconsistent with the model. Because ARP/wARP involves updating on the basis of density maps, such discontinuity can lead partially to slow convergence or even non-interpretable maps.
A very common misconception is that you have to have experimental phases to that resolution. No! Though it would be advantageous, it is not at all a requirement. You have to have native data to high resolution, do a quick extension to around 2.5 Å by solvent flattening, and then go on with ARP/wARP .
In general, the number of X-ray reflections should be at least 6 times higher than the number of atoms in the model.