I had to give up on both: they seem great until your page needs more than 10 formulae and it starts to take quite a lot of seconds for your page to finish loading for folks. Instead I switched to preprocessing my pages to extract latexy blocks, literally run them through latex and pdf2svg to turn them into clean SVG image and replace the on-page latex code with presized <svg> elements: now your content doesn't jump around while loading (because the browser already knows how much space the image will take up), everyone sees the same thing (no browser quirks around MathML etc) and because it's SVG, positioning and scaling is trivial.
I could trivially automate it so that it runs as part of CI/CD, but that's honestly more work than just running `npm run build` ever time I write a new chapter.
