Computed speed index far beyond reality in some cases

See this static HTML test file which fills any screen (above the fold) by delivery of the first 14 kB:
https://netzklad.de/test.html

It contains at its bottom a very large inline image. I prepared this months ago to prove that the first 14 kB **matter for speed perception** - and that browsers render to screens even if the rest of a HTML file is still not downloaded yet. That worked perfectly and shows what browsers do; see WPT waterfall:


Now look at **what Lighthouse computes** for this case, both the same in PageSpeed Insights and in Chrome:


This is a huge difference compared to reality. For me it looks like a bug. Test it yourself. After receiving first data the page is interactive instantly (scrollable in this case), the speed index even on crappy mobile phones is far below 5 sec's, JS is not involved.
The score of the old Google Insights version (before the Lighthouse update in November was applied) showed for this scenario 100 points. And this had met reality.