Research update: Towards a Law of Iterated Expectations for Heuristic Estimators — LessWrong