For example, in the design of a telecommunication network, the problem description might demand that every pair of nodes in the network must have at least two paths between those nodes for robustness and reliability.

To meet that demand, we'll design a model, taking into account that constraint on every pair of nodes. Let's say we implement a first-cut of our model, and Solver finds a solution for us--a solution that is quite logical: it consists of a huge loop, passing through every node. When we show this solution to our clients, we learn, alas, that solution is not acceptable in practice, so we go back to the phase of describing our problem again, and we add constraints on the length of paths between pairs of nodes. In consequence of this change in the problem description, we need to change the model and the implementation.

A better problem description in the first place--one that included all the constraints, especially those on the lengths of paths between pairs of nodes--might have spared us this repetition. Certainly a sound initial problem description is the surest safeguard against the infamous encroachments of an incremental specification.