OK, just to update others with this problem, I sent a recording of the issue to the helpdesk, and they have forwarded it to the developers. As far as I can tell, there is some confusion about how it is *supposed* to work. In one post above, it is suggested by eventide that turning feedback to zero should result in only one repeat, but later the same poster from eventide says that the randomness in the behaviour is the way it is supposed to work.
If some of you others could try it out and see if you have more predictable results, that would be helpful. If you find it hard to work with in its current state, it would also be useful to post that here. As always, the eventide folks are being responsive, which is great.