This story has moved to /story/kcsat-ml-probing-reasoning-models-with-nationwide-cohort-human-difficulty/.