verify rationality by observing the production of novel results. They don't have to be Nobel-prize grade

I would expect even better results when monitoring catastrophes among the trained and the untrained. Things like "not going bankrupt" relative to those in a good control group, rather than "got tenure."