>>> "Edward Berry" <[email protected]> 08/04/20 2:00 PM >>>
I guess the problem is that the model has to be somewhat flexible, so tying them all to the same reference wouldn't work unless the restraints are very loose.

You might try coordinate ncs, breaking it up into several individual ncs_groups. If you don't know where the flexion is occurring, make a large number of small groups along the chain - maybe 5 or ten residues ar a time. After refining with this, superimpose using each group in turn, and see what other groups are simultaneously superimposed well. All those have the same operator, so can be combined into a single group.