Great to hear, Jono! I look forward to seeing your entry play. Well done on the speed: OpenTafl's built-in AI can be either fast or memory-efficient, but not both, and right now, it's more on the memory-efficient side. (To a degree. It's all fairly heavyweight, since I'm using the same code as for the engine.) I think the best I ever did was 200-300k nodes per second on fairly beefy hardware.
I'm fascinated by your results w.r.t. old-school technique against Monte Carlo. My gut suggests that MCTS will produce better results in the long run, although you're probably right that there isn't nearly enough expert knowledge, nor a large enough corpus of games, to make it workable. Maybe later on in the fall, I'll ask for game records from Aage Nielsen's website—there have been a few thousand to maybe a few tens of thousands played there. It would be a handy thing to have.
Hydroque, if you want to look at another, less well organized approach, OpenTafl's built-in AI package is here
. Start with AiWorkspace.explore.
Anyway, OpenTafl is now up to v0.3.3.2b, which is I-swear-for-real the v0.3.x stable release, barring any bugfixes required. Network play is feature-complete, including loading saved games. v0.4.x is upcoming, with two things solidly in line: variations during replays, along with support for puzzle-type saved games, and AI improvements. There are lots of heuristics known for chess programming I'd like to try to apply here.