My cool theory about machine-readable HTML

I believe that it's silly to publish markdown versions of your web page separate from the HTML ones, because it implies that your HTML is not accessible but your markdown is. If that's the case, why isn't your HTML accessible instead?

If you object that your HTML needs to contain a lot of things that aren't relevant but which browsers still know how to handle, why can't a machine process the HTML in a deterministic way to extract the semantics? In many workflows you're already doing that on the publishing side, so it must be possible to do it on the consumer's side as well.