Help improve reasoning evaluation in intelligence organisations — LessWrong