> Really we should have some kind of open source simulator of a market, and the surveillance state perspective of it, so we can prove in practice what is possible to deduce...

A variation of https://en.wikipedia.org/wiki/Random_forest might get the job done (the risk is an overfitted model).