[ewilliams]

Validation requires ongoing and in-depth interaction and communication with the client engineers and managers; it is a much more generally interactive process than verification. If the system being modeled already exists, clearly model results should be compared with current system reality. If the system being modeled does not yet exist, compare model results with a similar system if one is available. Model output should also be checked against expert opinion whenever possible. A particularly valuable validation method in many situations is the Turing Test (named after Alan Turing). In this test, the client is shown model output and output metrics from the existing system -- but not told which is which. If the client cannot tell them apart, the model passes the Turing test. If the client can, the model fails the Turing test, and the modeler next asks "How could you tell?" Example reply: My department always has a big backup behind the milling machine, and the model doesn't." The modeler now has concrete information and clues with which to find and correct the modeling error(s).