I think the first is better, because "failover" is a much more clear term than "contingency plan". Contingency plans could involve things that you just aren't willing to do in non-crisis situations (nuke Russia, or something), and making that simulated rather than real may undermine the value of the test. If you don't know what happens if you really nuke Russia, and you don't want to do it, making up a mythical "so now we have nuked Russia and X happens" may be contrived and inaccurate. Some things just can't be tested well.