I'd be curious if there were some measurements of the final effects, since presumably models wont <think> in caveman speak nor code like that