I am very wary of 'modeling' questions on MathOverflow. The statistics section is plagued by questions like "how should I represent similarity for my clustering?" I think we should strongly consider closing any question which demands non-trivial interpretation or modeling of non-mathematical statements.

This criterion applies to question 22490. In my answer, I had to assume Herb wanted to learn a causal DAG from observational data. I think the only reason the question is still around is because causation vs. correlation is a very important topic and most people are surprised to learn that it can be mathematically handled.