A key element of Cassandra's design is the use of a single format to represent all sources of uncertainty that affect planning. In particular, all uncertainty is assumed to be manifest in uncertain effects of planning operators, as outlined above. Uncertainty about initial conditions can be handled within this format by treating initial conditions as though they were the effects of a phantom ``start step'' action. This treatment of initial conditions, which was initially developed for reasons unrelated to the problem of representing uncertain outcomes, is common to the SNLP family of planners to which Cassandra belongs.
Cassandra's formulation ignores uncertainty that might stem from outside interference during the execution of the agent's plans, except inasmuch as it can be represented as incomplete knowledge of initial conditions. This is, of course, a limitation of classical planners in general; all change in the world is assumed to be caused directly by the actions of the agent.