I believe you, obviously. I have historically received drastically different results between, for example, web.dev / Lighthouse and PageSpeed Insights with the latter outputting a very low arbitrary 'rating' or 'score' and the former providing results I would expect. Like a lot of efforts at Google, things seem duplicated. As an outsider, it is a frustrating experience catering to Google's performance ideals because the tools and 'best practice' positions do not seem to be unified in one place.
I wonder if you might have had a bad experience with the old rules-based PageSpeed Insights? It wasn't as good as the current one, being more of a checklist of "have you done all these things that usually help" and not "we loaded the page, computed metrics, and this is how well it did".
The previous version of PageSpeed Insights was open source (https://github.com/pagespeed/page-speed) and the current version is a wrapper around Lighthouse which is also open source (https://github.com/GoogleChrome/lighthouse).
(Disclosure: I work for Google)