Inferential credit history — LessWrong