(oh, also: yes, that's my understanding too, that you're still stuck validating incoming curve points in ECDH, but I think the anti-cofactor people have a corresponding argument for signature schemes.)