I've managed to build a working(*) #retro 12 for 16-bit 8086. It fits (barely) into a single 64K memory segment. It's also *very* slow, as one might expect for a 32-bit virtual machine running on a 16-bit architecture..
The next step is to implement a virtual memory system so I can have more than 10k cells (the standard image + input buffer uses 8,575 cells).
(*) strings aren't working fully due to insufficient space for the rotating temp string buffers.