The best method for computational structural validation is trying different methods and seeing that they return similar results. I’ve heard good things about the Zhang Server/I-TASSER.
RMSD cutoff for clustering is dependent on the system, but you want a number that give you a manually manageable number of clusters without allowing the clusters to vary too much internally. I think realistic numbers are in the 1-10 sq angstrom range but that’s a guess.
The codebase was branched for 3.2 but I don’t know how far along the validation process is before we can release it. There is a specific plan for 3.3 in mind tied to certain publications; at this point 3.2 may get delayed and we’ll merge 3.3 into 3.2 (which would delay 3.2 until February or so). The lab managing the release is the lab that wrote flexpepdock so you can always try asking them directly.