The why is that CAESAR runs several numerical simulations on the model in order to error-check itself, and that the method in question is resulting in a non-linear support switching between "in use" and not "in use." E.G. in odd-numbered calculations, CAESAR thinks a given support is sliding or hitting a guide (or whatever non-linear condition), and in even-numbered calculations, CAESAR thinks the same support isn't slighting or hitting a guide (or whatever non-linear condition).
The strange part is that this can be caused by a number of non-linear supports elsewhere in the model culminating and centralizing here.
Verify that your VC CNODE supports make sense, and aren't too elaborate. Then look at other supports in the model, particularly near your VC. If any are lifting off (or are trying to), you can adjust the friction value (or removing it).
If you still can't get it, try posting a screenshot.