Single Comment

Yes, of course it is very weak. It searches only 2 ply, and that isn't even enough to see a mate-in-1 threat coming. You can switch it to 3 ply, but in a game as large as Chu Shogi that would probably already make it think too long. This is JavaScript, not a C program!

I am sure I could make a search that plays much stronger with the same number of nodes. I put this one together in two days. But that is not meant as an excuse: I intend this thing to be weak.  It is meant to be a demo for people that have never played the game before, and visit the page the diagram is on to learn about its rules, so they can have a sparring partner to get a whif of how the rules work out in practice.

Also note that it doesn't use any variant-specific knowledge other than the rules. It has to guess the piece values itself, and I am amazed of how reasonable these guesses appear to be. I made a version that also tries to weight in promotability, by making some weighted average of the values of base piece and promoted form, where the weight of the latter increases with the ease of its promotion and the relative gain of intrinsic value. That also looks reasonable, except that the current algorithm then puts an unexpectedly high value on the Lance. It seems this has the most favorable prospects for promotion, so that it assumes you would do your best to preserve it until it can actually promote, rather than lose it in tactics to clear the way for others. (Of course it doesn't realize the Lance is file bound, so that you really don't have much choice in the matter...). But now that I think about it, perhaps this is not such a silly thought. The Lance itself is pretty useless in combat, can reach the zone in a single move when the time comes, and a White Horse is one of the more valuable pieces.