However the consensus guess is that perfect play yields a draw.
Just like we can efficiently find approximate solutions for the traveling salesman problem (that are at most 50% longer than the optimal solution), these heuristics have not much to do with the optimal solution.
> I could imagine that a perfect play completely contradicts common chess theory.
So can anyone. Nobody knows what perfect chess play is. Our best guess though is that whatever your opponent plays you can always force a draw before they can win.
Unfortunately it leaves us in the position of not having much to go by, if expert experience doesn't help analyze the game.
I wonder how different a chess engine and optimal play look for the reduced sized boards.