The API reference by category of the SDL2 wiki is still a mess, arguably to the point of being worse than nothing since it lacks indication of how incomplete it is and people coming to libsdl.org are pointed toward it as the main documentation rather than the header files. Did you ever run the rebuild on the SDL2 wiki, or only SDL3?
Compare the current category page for SDL_video.h with the equivalent page from the archive of the old wiki:
https://web.archive.org/web/20210210032231/http://wiki.libsdl.org/CategoryVideo
The new wiki page is a list of 19 links in alphabetical order, with enums, structs, and functions jumbled together and no indication of which each link is. The old wiki page has separate lists of 10 enums, 6 structs, and 86 functions.
While we’re at it, compare the SDL3 version with the old wiki as well:
I assume it’s complete, but It’d be nice if the enums, structs, and functions were separated from each other, and the links weren’t double-spaced (maybe double-columned so you can get the whole list on screen at once, currently it’s a 4-page scroll on my 1080p monitor).
Subcategories smaller than header files would be nice. You could have a cloud of words appearing in the names (Window, GL, Get, etc.), and filter by selected word(s).
As for “you can click the ‘Edit’ button at the bottom of any wiki page to make changes, as one expects a wiki to work”, this is not really the case. The edit button lets you log into github to submit a pull request to a repo through a wiki-editing-like interface, and some of those pull requests have been sitting open for over two years. This is not how a wiki works, which is why obvious and severe problems are sitting on the “wiki” for years without being fixed by users who are bothered by them. Wikis recruit effort with low friction and immediacy of results. This edit button is more of a suggestion box.
Anyway, I’m not sure there’s any benefit to having the API reference be published as wiki content. It could probably be both a lot nicer and easier to maintain as a set of static pages automatically generated from the header files (which have comments containing information that’s very complete, but also clearly organized to be machine-processed, to the detriment of human readability).