Functional programs that explain their work

We present techniques that enable higher-order functional computations to "explain" their work by answering questions about how parts of their output were calculated. As explanations, we consider the traditional notion of program slices, which we show can be inadequate, and propose a new notion: trace slices. We present techniques for specifying flexible and rich slicing criteria based on partial expressions, parts of which have been replaced by holes.