I think I understood you to say that the only way to really verify whether the modelling being done bears out is to build it, monitor it, and then assess whether it matches the model. Is that correct?
On June 16th, 2009. See this statement in context.