Hi Gerwald,
Two more quick questions: how would you identify the best strategy?
I would look at Rfree and Rfree-Rwork, and maps of course. To help in deciding what are the "good / reasonable" values, use POLYGON (Acta Cryst. D65, 297-300 (2009)) (phenix.polygon), local real-space correlation (phenix.real_space_correlation). I think Nat bundled them all into his magic tool called phenix.validation with will do this and more (it is GUI based).
I would aim for the lowest number of parameters with reasonably low Rfree?
Yes, this sounds right in general.
Does phenix report the total number of parameters refined anywhere in the log?
It's trivial and not at the same time (at least the way I understand it, of course I might be very wrong). It's trivial to compute the number of refinable parameters given the strategy you choose, but how you will count the restraints which are kind of quasi-observables, and what about weight that if chosen too strong so it can turn the restraints into quasi-constraints? Pavel.