is there some kind of general trend if not consensus as how to
refine cryo-EM structures?
Not really consensus, but my clearly biased contribution to the topic:
I'd say basic common sense for refinement applies:
What do people do for maps in the 3-4Å
range,
3-4A may require secondary structure and rotamer restraints. Use "NCS" if available. If map is symmetrized, then use "NCS" constraints, otherwise restraints should be just fine. Typically, data at this resolution range does not allow to confirm geometry validation outliers, so your aim is no outleirs. Etc, etc. And again, validate the model and model-to-map fit, locally (at atom/residue level) and globally (standard "Table 1" metrics, discussed here
https://doi.org/10.1101/279844 , for example).
pure RSR against the untouched map and that's it?
Good idea, in my opinion.
Reciprocal
space refinement against the backtransformed structure factors
from the map but without touching the map?
Personally don't like this idea.
I've even heard that some people "dare" to calculate
2mFo-DFc-maps....
Not a good idea, I'm pretty sure. Cryo-EM map is not biased by atomic model (normally, unless a model was used in reconstruction). 2mFo-DFc map is model biased by definition. Also needless to say that absence of "Fobs" in cryo-EM is a good hint that this map isn't what you want -;)
All the best,
Pavel