Preference framework — LessWrong