Fun think about core web vitals and amp: AMP pages will _always_ win core web vitals, because they're measured from google's CDN, where they can be pre-rendered. Unless the page is motherfuckingwebsite.com, the AMP version is going to score better in every metric than a non-AMP version. Core Web Vitals just cements AMP as a shortcut to performance.
Have you directly compared the live version against that in the AMP Cache?
The idea that PSI preloads the page before starting the test doesn't make much sense at all, and doesn't mesh with my own experiences (in the same space).
