Troll vs Castles

Posted on 2020-07-07

Troll vs Castles (commonly abbreviated as TVC by now) is a simple multiplayer game I’ve been working on for a bit of time now.

As it succintly mentions in the statement, I learned about it on the #Fr channel from @Zaap38, who got it as a short-term assignment from a professor of his, who took the idea from another professor acquaintance of his, or so we suppose. In the end it leads back to this document {pdf,fr} by Romain André-Lovichi, who’s apparently been using it as an AI introduction lab exercice for his Terminale (~17yo) students.

That game is close to perfect. It’s easy to understand, it does not have a single one-size-fits-all strategy, it doesn’t depend on luck too much. Too much.

It’s a simple balance between gaining a positional edge that awards immediate victory, and an energetical one that awards freedom of movement, and deferred loss if you run out of it.

Its only weakness is that it needs some repetition to reliably observe a better player. (I’m currently working on mitigating that.)

Truth be told, as a perfect information symmetrical zero-sum game, there necessarily exists a closed-form optimal solution.

I’m not going to spoil it too much here. It’s neither trivial to come up with, nor to fit in the code size limits, nor to fit in the TrueSkill-induced meta.

So go ahead and try it out!

Very simple strategies are possible, and can easily reach the top ranks. Whether they can reliably stay there is another question :-)