pandoc is much less powerful here since it can only parse a subset of LaTeX, as opposed to LaTeXML or lwarp which actually processes the LaTeX/TeX code expanding the macros etc, and then generates the output HTML/XML.
pandoc is nice for simple documents, but it is completely inadequate for many use-cases, which is why ArXiv is using LaTeXML and tikz.dev is using lwarp, rather than the more ubiquitous pandoc.
I’m especially interested in how this approaches page flow - would it keep the concept, or render everything as a single “page”.
For context, in my pet project I wanted to actually keep the pages; I ended up rendering a pdf and then using pdf.js to render that in html.