Hello,
I guess it is not terribly useful in most cases, but it would be a valuable tool for comparing the outcomes of different refinement strategies for the same macromolecule, especially as rigid-bond-test, unlike ADP values, should be in principle independent from effects of different anisotropic scaling factors.
Averaging rigid-bond-test values over domains, main chain/side chain of a certain residue, or over certain bond-types (peptide, C-Cring a.s.o) would be a nice validation tool too.
In case of isotropic atoms it's just |Bi-Bj|, where i and j are indices of bonded atoms, B are their isotropic B-factors. Often one considers <|Bi-Bj|>, where <> is average over all such pairs (or selected atoms), and I would expect as many opinions about what these values should be as many people are aware of such metric. In case of anisotropic atoms the formula is the same, but B means projection of anisotropic vibration component onto bond vector i-j. Hirshfeld's paper (Acta Cryst 197?, ?=6..7...8) discusses values for |Bi-Bj|, but it isn't clear whether that is transferable to macromolecules. If memory serves correctly, Dale Tronrud uses tabulated |Bi-Bj| values for various atom types derived from well determined structures as targets for ADP restraints. Some programs explicitly apply it as part of ADP restraints. Pavel