Formalizing reflective inconsistency — LessWrong