Given the variation between results for different, but similar capability models it is hard to take this seriously