for me at least i had mentioned it offhand a couple times but never posted about it as a dedicated topic because i (obviously in hindsight very incorrectly!) assumed that everyone who was familiar with llms would've realized it by then. your mention of the spiralbench penalty and then the reaction to your post and lindsey's paper really took me by surprise and tbh updated me pretty negatively. (and my current research direction is actually trying to formally demonstrate something similarly "obvious" because of this.)