Upon reception of a high-level goal from FORMAT, Prodigy both searches its case library for appropriate cases, and simultaneously generates a new plan for the high-level goal. Prodigy searches for cases to make the returned plans be as similar to previous plans as possible. This similarity makes the detailed planning and execution of those plans simpler.
The reason to generate a plan in conjunction to searching the case library is to ensure that all the conditions for the cases have been checked. In particular, if a piece of planning knowledge would affect the plan only when certain conditions exist, then a case might have an over-general index with regards to those conditions. As an example, if dogs are required to check for drugs when securing an airport in a drug smuggling area, then a case for such a plan would both test if the airport is in a drug smuggling area, and would include the shipping of dogs. However, a case for securing an airport in a non-drug smuggling area might not include the negated test (NOT drug-smuggling area). Not including the test makes the case index over-general. Similar results can be obtained if the knowledge base used to create plans is increased over the knowledge base used to create the stored cases. Thus, the generative plan is useful as a check on the validity of the cases.
The feasibility of a plan could be checked at various times during the planning process. We have chosen to check the feasibility only when a complete plan has been created. This allows both the cases and the generated plan to be checked for feasibility. Thus Prodigy passes a complete plan, along with any constraints between plan actions, to Ditops. Ditops then takes the plan and attempts to schedule it in conjunction with all the previous plans.