Lighthouse does a 4x slowdown for mobile. It does not do a variable amount of slowdown based on the speed of your device. Therefore it is not a calibrated slowdown—running Lighthouse on different devices will result in different results.
If you hover over the "Emulated..." setting you'll see an estimate of the base power (447 in this case). Roughly you can assume this run is therefore approximately 447/4 = 112. I'd imagine your developer PC is much more powerful than the machines PSI uses and this explains the difference.
My M1 Mac gets a score of 88 with a base power of 2842 (or 2842/4 = 711 after throttling) so 7 times as powerful as the PSI machines! No wonder there's quite a difference!
So why doesn't Lighthouse calibrate properly to give a consistent score no matter where it's run? Well that's actually incredibly difficult. Particularly on machines being used for other work (light your developer PC). You can
calibrate it better when running Lighthouse separately. But also remember the more you calibrate, then the more you're kind of fudging the results.
At the end of the day, Lighthouse is just one example of what performance would look like on a site and, more importantly, gives advice on how it could be improved. The PSI machines ARE quite slow, but that's OK as they are intended to show potential issues as experienced by real users (who are often on a lot slower devices than us developers!). Although it's tempting, we really don't recommend assuming the Lighthouse score is a definitive measure of your performance. It's a high-level summary of that specific run. It's best used to compare two runs (e.g. before and after a fix) run under the same conditions.
The "Discover what your real users are experiencing" section is more important as a measure of how your real users are experiencing your site (in all their various devices and conditions!). Use that as the measure, and use Lighthouse as advice on how to improve it. The Lighthouse score is useful however to check that run against the real user data—if real users have issues not shown by the Lighthouse summary, then its advice is not going to be that useful. Similarly if real user scores are great, but Lighthouse is showing it as poor, then you know the Lighthouse audits are areas of potential improvement, but it maybe isn't as urgent to address since the real users seem to be having good experiences.
I hope that helps to explain the differences you're seeing, and how the Lighthouse results are often still really useful despite that. In fact the slower scores, as given by the more constrained PSI devices are often MORE useful as they highlight more opportunities!