It seems that the right way to handle a malformed route or two depends on who's speaking and who's listening. If I'm a backbone provider and I hear a bad route from a customer, I'm going to drop that connection. I have no incentive to take any risks. This is just as the RFC currently reads. If I'm a customer, I really don't want to shut off the service that I'm paying for. If I'm not going to propagate the routes beyond my borders, why should I drop the whole session? The risk is entirely mine, and a partially corrupted table is better than no connectivity at all.
From this point of view, it seems that the RFC should be loosened to allow configuration of a BGP peer to continue the session and ignore the route.
Perhaps there should be wording to the effect that it is not acceptable practice to propagate routes from the offending router beyond your borders. Maybe there is even a way to phrase it that means "it's not OK to propagate routes from a suspect router back into the core of the Internet." In practice these words have meaning because "upstream" and "downstream" are defined by the flow of money, and economics suppresses loops. Steve Schaefer Dashbit - The Leader In Internet Topology www.dashbit.com www.traceloop.com