The funny thing is that HTML was supposed to be a markup language that could be read/written by hand, while making it also machine-to-machine friendly - notably by making some "semantic" features accessible for browsers. One of these for instance is the structure of the document; marking section headers was supposed to let browser to automatically generate a table of contents. Additionally CSS was supposed to let users choose how all this was displayed.

All of this failed - or rather, was undone and cancelled by the "modernization" of the Web. Namely the arrival of for-profit companies on the Web, be it Facebook of the press like the New York Times.

It was a boon as they brought valuable content, but they brought it with their own rules. The first set of which is the ads-supported model, which is by definition the opposite of free content; an ad-supported website is not free in a very sneaky way, and it's not just about privacy and manipulative practices (targeted ads, as if ads were not already manipulative enough). Users are actively prevented from consuming the content the way the want.

The situation today is that very few browsers offer out-of-the-box a way to apply a personal CSS, and I think none will generate a ToC from the headers of a HTML page.

And the "semantic" part - far from specialized and more accurate semantic markup frameworks that were considered - is being completely taken over by LLMs; an insanely expensive brute-force solution IMHO.

The web has already be reinvented mostly the way you suggest, see for instance the Gopher and Gemini protocols, but they'll stay forever "niche" networks. Which could be not so bad, as it is very clear that the Web is full of actors malicious to various degrees. Tranquility by obscurity?