Report likelihood ratios — LessWrong