If the rollover of a vehicle is a stochastic process, what hope is there of predicting all the details? Worse, if all one has to go by is the evidence remaining a few years later, how can the details be reconstructed? In a 1989 paper,1 it was argued that a computer simulation adapted from an occupant model could be used to replicate (after very considerable effort and “tuning,” no doubt) the vehicle motions seen on video during an actual rollover test, if only for the first second or so. Even if the simulation effort were successful, one can virtually guarantee that it would not predict the motions of an identical vehicle, if an identical test were to be run the following day. Claims that a computer model can “be an effective tool in reconstructing the motion of the vehicle”2 would seem to be preposterous on their face.