You may see some older discussion of the
accessibility was developed and published back in 2007 as a proof of concept for some of the KOMA document styles. I got hold of the files from the author in 2019 and took over maintenance with her permission. I tidied up the package enough to get it to CTAN, but didn't update the functionality. I also published it to GitHub to get some feedback on it.
It seems to have worked well in 2007 for a few test cases. Unfortunately it now fails every test case, and it looks like needing some serious efforts to fix.
Because of this I no longer think that
accessibility is fit for purpose. As you've seen I've recently asked CTAN to add a disclaimer to the catalog entry for it. The Github code repository is also tagged as "not ready for production". So, don't try it unless you are curious! If anyone reading has coding skills and would like to contribute to the package, please leave an issue there (or, just fix it...).
There's another issue here. Tagging and accessibility is deeply linked to the semantic structure of the document. There's a need to separate the the visible structure in the output PDF (that which a sighted person would recognise from heading numbers and good use of visual clues) from the document content (visible as the tagging tree in adobe, for example) so that machine readers can use it. Ideally this would be done within the core latex code so that a solution is usable by every package, not as a sticking-plaster on top where every change to a package breaks it.
This need for a broadly applicable solution is why the approach being pursued by @Ulrike and the core team is much more likely to lead to a sustainable solution. Also there's stuff to be rescued from
accessibility, there's some great ideas in
tagpdf, and the whole thing has to not break the rest of 'tex. Not much to ask! So.. currently I'm thinking on ways to get some resources for this and will be coordinating with the team. All suggestions are welcome!
hyperxmpis the most recent and sophisticated. In terms of PDF/A, conformance «a» (but not PDF/UA) also colour management is a requirement. Right now, the necessary OutputIntent is best included manually, see an example. Since
tagpdfis rather labour intensive you may consider the Acrobat approach.