So, a megabyte of PDF which both downloads and renders far more slowly than the equivalent HTML, which could be 20KB of HTML, no JavaScript, 1KB of CSS and perhaps 50KB of images (which don’t block page rendering, either), all bundled into one HTML file (base64 the raster images) of under 100KB.

I can imagine why you might have concluded as you have, but your conclusion is nonetheless baffling. Most of the reasons you’ve chosen PDF over HTML apply just as truly to HTML; page-orientation is the only one that doesn’t, and I refute your claims that that page orientation is a good thing for the web at large or the hardware that most use. (Guess how I read the document? By scrolling, not by pagination; the footer and header of each page transition is just a minor annoyance in the middle of a paragraph of text.) I disagree with every point of your assessment on PDF’s historical disadvantages, too: PDF as used still includes patent-encumbered things, and implementations are complex enough that major deviations and incompatibilities are common; PDF files are still ginormous (nothing has changed about this ever), and tooling is terrible (perhaps even non-existent?) for trying to shrink files without butchering the lot, with yours as a good example of not being an order of magnitude smaller as claimed; it’s still far harder to make PDFs accessible, because most free tools just can’t do it, and it takes far more effort than HTML where it’s easy (as before, tooling is terrible, where with HTML you can just edit the source in a text editor); most PDF readers can’t reflow text (don’t think I’ve ever used a tool—reader or writer—that could); most PDFs can’t be edited freely; and PDFs don’t render well on screen.

You’ve thrown out the HTML ecosystem because some people abuse it, and are choosing to use PDF despite the rampant abuse and many problems of the format, because it’s theoretically possible to work around those problems. (You may quibble with my judgement of “rampant abuse”, but most PDFs I encounter on the web are PDFs for no good reason and perform terribly, loading slowly and being worse for task performance. It is perhaps a slightly different class of abuse, since spying on users via JavaScript isn’t part of it, but it’s related inasmuch as it’s not about meeting the needs of the user.) This does not seem internally consistent.