Hi,
Sharpening B-factor estimates from the Guinier plot are unreliable at low resolution (less than ~7Å). In NU-refine and homogeneous refinement, this quite often leads to bananas B-factor estimates (1000-2000) for low res reconstructions (resulting in a sharpened map that has essentially fallen apart into noise).
This is not really a problem (the half maps are there, after all) but I wonder if it would be worth defaulting to a fixed B-factor until the GS-FSC exceeds some threshold value, or the fit to the Guinier plot exceeds some criterion of reliability.
Cheers
Oli