I have done software screen mirroring for RTL languages on an Atari ST ages ago. There I had both full images available and I'd just compare source and target r

I have done software screen mirroring for RTL languages on an Atari ST ages ago. There I had both full images available and I'd just compare source and target rows, skipping 8 rows (IIRC) until detecting change, in which case I'd swap the local area. The start row was incremented after every scan, so I'd be sure to not have missed any pixel after 8 scans. This allowed to track the mouse quickly and also to respond to local changes (text editing) without letting the mirroring thread fully lock up the computer. For your case, it seems more reasonable to divide the screen into some kind of rectangular cells, keep checksums for the cells and switch them to A16 if the checksum stays constant for two visits? This would avoid the need for a full extra frame buffer while still being relatively accurate. I think one main problem will be that you'd need to modify the screen firmware to do something like that. Probably would make sense to code a software emulation on a regular screen to see what it would look like...