We discuss the PID-abductive framework, in which explanations are selected depending on how they are used. The framework has been formulated with plan recognition as its target application, but is generally applicable to abductive explanation. The framework addresses in particular two problems for explanations: The selection of the best explanation, and the issue of selecting what to explain, and what should be regarded as facts. The PID-abductive framework is characterized by the fact that observations are viewed neither as facts nor as things to prove, but as conditions on proofs for desired actions. Intuitively, this means that we will seek an explanation for an observed action, only if this explanation is important for the selection of a response action. This requires a deduction system which allows one to reason "backwards" from conditions to equate a condition with a disjunction over its possible explanations. A candidate proof system is Partial Inductive Definitions, which contains a deduction rule Definitional Reflexion that does precisely this. Using Partial Inductive Definitions also achieves that assumptions typically rule out possible explanations of observations, allowing for more adequate minimisation criteria. Much work remain. Most serious is the task of defining minimisation criteria, and a search strategy that does not exhaustively explore the search space of all possible explanations.