RAISE post-mortem — LessWrong