It is possible that a single plan could work just as well for several different outcomes of an uncertainty. For instance, suppose the action of asking Chris for $50 has three possible outcomes: either Pat gets the money and Chris is happy (at having had the opportunity to do a favor); or Pat gets the money and Chris is unhappy (at having been obliged to do a favor); or Pat does not get the money at all. If Pat constructs a plan in which she tries to borrow $50 from Chris to bet on a horse, then, assuming that this plan does not depend upon Chris's happiness (which it might, for example, if Pat needed to get a ride to the track from Chris), the plan will work for either the ``get money + Chris happy'' outcome or the ``get money + Chris unhappy'' outcome.
Cassandra could find such a plan, but would in effect have to find it twice---once for each outcome of the uncertainty---and it would still require a decision-step to discriminate between those outcomes. This is inefficient in two ways: the extra search time required to find what is essentially the same plan twice is wasted, and effort is put into making an unnecessary decision. We are looking into ways to avoid the former problem. The latter could be solved by a post-processor that would ``merge'' identical contingency plans, but we have not implemented this technique.