My understanding was that the effort level settings on models is essentially throwing more compute at CoT, and it's trivial to see that more inference-time comp

My understanding was that the effort level settings on models is essentially throwing more compute at CoT, and it's trivial to see that more inference-time compute on a task results in better results. Is that not accurate? Because if so, I'm unclear on how CoT could be characterized as performative.