I've always wondered whether there's a bunch of conditional statements constraining the output of the probabilistic model. That would seem like the logical thing to do, but I'm not that familiar with ML to know whether such thing is needed or not.