I did not exactly explain the effect, just pointing out that very simple statistical measures are sufficient to pick out unusual text, and some kind of "awareness of being tested" is not necessarily indicated

→ View original post on X — @plinz