Indeed, this is a not-always-wanted consequence of configuring the Diagram with stalemate=win. Which is needed in games without checking rule to suppress declaring a draw when the all pseudo-legal moves would expose the King to capture on the next ply (if the current ply did not finish the game). In the AI this is implemented by awarding moves that annihilate enemy royalty (and not own royalty) the highest possible score ('+INFINITY'), so that the preceding ply that allows this to happen receives score -INFINITY from the POV of the player then on move, which can be interpreted as 'illegal'. As this is also the initial value of the 'bestScore' variable for that position, this means that bestScore is never incremented if there are only illegal moves. But also if there were no moves at all, and the cases are not distinguished.
This could be prevented by assigning moves that annihilate enemy royalty a score of INFINITY-1 in games that have no checking rule. Then the moves that would otherwise have been illegal get a score -INFINITY+1, and if there are such immediately losing moves, bestScore would be incremented to that (but still not considered a very good score). Since the stalemate test (being in check, i.e. losing the King after turn pass) would only be executed if bestScore has stayed at -INFINITY, to correct that score to 0 (the draw score) if there was no check and stalemate=draw. That would then still apply to these deadlocks. An extra parameter could be added to the Diagram for deciding whether royalty extinction should be awarde INFINITY (checking rule) or INFINITY-1 (no checking rule).
I must say I don't worry about this very much. The design goal of the Diagram is to provide a sparring partner for the opening and middle-game of a variant, not as an arbiter on the rules in positions one would never encounter in a real game.
Indeed, this is a not-always-wanted consequence of configuring the Diagram with stalemate=win. Which is needed in games without checking rule to suppress declaring a draw when the all pseudo-legal moves would expose the King to capture on the next ply (if the current ply did not finish the game). In the AI this is implemented by awarding moves that annihilate enemy royalty (and not own royalty) the highest possible score ('+INFINITY'), so that the preceding ply that allows this to happen receives score -INFINITY from the POV of the player then on move, which can be interpreted as 'illegal'. As this is also the initial value of the 'bestScore' variable for that position, this means that bestScore is never incremented if there are only illegal moves. But also if there were no moves at all, and the cases are not distinguished.
This could be prevented by assigning moves that annihilate enemy royalty a score of INFINITY-1 in games that have no checking rule. Then the moves that would otherwise have been illegal get a score -INFINITY+1, and if there are such immediately losing moves, bestScore would be incremented to that (but still not considered a very good score). Since the stalemate test (being in check, i.e. losing the King after turn pass) would only be executed if bestScore has stayed at -INFINITY, to correct that score to 0 (the draw score) if there was no check and stalemate=draw. That would then still apply to these deadlocks. An extra parameter could be added to the Diagram for deciding whether royalty extinction should be awarde INFINITY (checking rule) or INFINITY-1 (no checking rule).
I must say I don't worry about this very much. The design goal of the Diagram is to provide a sparring partner for the opening and middle-game of a variant, not as an arbiter on the rules in positions one would never encounter in a real game.