Alternatively, our screen readers, particularly speech synthesizers and Braille translators, should be smarter. espeak-ng is already pretty good at splitting all-lowercase compound words. I'm not one who believes that the sighted majority should all have to put in extra effort to accommodate us; our tools should do as much of the adapting as possible.

And yes, taken to its logical extreme, this means that screen readers should use machine learning to reconstruct UI content and semantics from pixels, rather than developers having to make their GUIs accessible). I think we'll eventually get there. But that's way harder than handling all-lowercase hashtags. And requiring all social media users to change their habits just for us is a much bigger ask than requiring application and GUI toolkit developers to implement accessibility.