A problem we have faced in the SLIE grant is how to describe the requirements of experiments on agent models, since often these are most conveniently expressed in terms of very large populations rather than as direct properties of individual agents. Part of the SLIE group (Wamberto plus our Barcelona colleagues) came up with a solution to this, which we honed a little (to the half-baked stage) at a joint meeting of all the grant partners. Since Wamberto currently is ill I shall (unkindly) steal his thunder by presenting the solution, which involves the use of models of continuous change for expressing requirements and validating agent-based models against those.