Still can't tree shake. It's what makes my old css files grow into monsters. I feel like everyone has given up on this.

For small-scale projects, I use a server-side helper for an HTML page to include only certain CSS, SVG, and JS on a page. Each JS file is tiny, progressive, vanilla JS; SVGs are predefined and generate an SVG sprite that I then consume; CSS pulls in a `base.css`, the page's CSS file, and any component ones pull in their own. It's hand-rolled per project (pre-AI being decent), but I'm proud of it, and it keeps every page tiny. HTTP/2 multiplexing is pretty good and downloads the files more or less in parallel.

CSS modules ought to make it possible. But it’s more or less impossible with regular CSS, you have no idea when a class/ID might be used.