A formalism is elaboration tolerant to the extent it is easy to change in order to reflect new information. Natural language is the model for elaboration tolerance; in everyday discourse one can easily and succinctly change one's declarations by adding an extra sentence or two. Elaboration tolerance is a desirable property of any representation, whether it is a logical theory, a Bayesian network, or a computer program.

We introduce a kind of elaboration tolerance that we call additive elaboration tolerance, where a representation is modified simply by adding the appropriate formula. This modality of changing representations is attractive because it avoids "brain surgery" -- we do not have to tinker explicitly with the representation at all. We show how to endow a given representation with additive elaboration tolerance by embedding it in another system. This embedding has some useful properties, such as modeling human discourse, and preserving the facts true in the original representation.

We conclude by relating some design principles that can promote elaboration tolerance in general. These principles are extracted from various fields of computer science, including relational database design and object-oriented programming. We show how various solutions to the missionaries and cannibals problem follow from these principles.