@Noino Hi! What you’re doing is great! We would like to do something similar with TZP (it even has some global JS variables to use for that, we never took the time to create something that uses them).
That is sorta known. The font fallback is async, and the second refresh will have everything in cache.
I would have to check the source code of the test to understand which glyphs they are testing.
But before I do, I might already know the cause: some distros have updated freetype and harfbuzz. Sadly, this is affecting how some glyphs are rendered, which might explain the change you see.
Do your docker image and configuration on desktop have the same versions of those libs?