So far I've managed to write the core of the engine, including the
load/save system, prototype map generator, and line-of-sight code. I
can happily wander around the map opening and closing doors.

http://www.phlamethrower.co.uk/misc2/batr1.png

But then I tried running it under a BBC emulator, and discovered how
hideously slow it all is. Some of it seems unavoidable (20 seconds to
load the 4KB data file with all the dungeon items in), other bits are
due to the unoptimised state of the code (taking 1 second to draw each
map cell), and other bits suggest I'm pushing the machine too hard -
the line of sight algorithm takes around a minute to run!

Admittedly I am only using a very simple ray casting algorithm (mainly
to keep code size down - it's already up to 10KB, out of 16KB max I
have available when taking all the other data into account), but even
if I were to take the time to implement a beam casting or similar
algorithm I doubt it would approach a playable speed. So currently I'm
trying to decide whether I should switch to a different LOS technique
(e.g. a room gains visibility when you enter it, but remains visible
even if you go elsewhere), or ignore the speed problem and have a go
at crafting some assembler optimisations once the competition is
over :)