Thanks for the link, this Kantu XClick and XMove image-driven "real user simulation" project opens up some interesting possibilities.

It also points out a significant limitation of Cloud Browser's imagemap-based architecture. Screenshotting the page locks away the text behind an image, inaccessible to screen readers, copy and paste, or other interactions. This doesn't seem easy to solve, since the most the img tag offers for accessibility is an "alt" tag, which does not allow specifying which areas of the image contain what text. Sending the actual text (like html.brow.sh) would solve this problem, but then layout is up to the end-user browser again.

Client-side OCR'ing of the image may be a possible alternative, will look into it thanks!