To expand my Go comment (did I say this somewhere else? I feel like I did but I can’t find it), what I mean is that you could generate possible configurations and rank them by their semantic content
For example, you could feed each possible configuration through GNU Go, asking it to score the configurations, and pick the configuration which is most ‘intelligent’/likely to be produced by strong scorers.
Or, you could instead tell GNU Go that ‘black is a 20 kyu player, white is 1 dan’ and rank configurations by which configuration is most consistent with such a differential (configurations with stupid moves by black being far more likely than stupid moves by white, and the converse).
To expand my Go comment (did I say this somewhere else? I feel like I did but I can’t find it), what I mean is that you could generate possible configurations and rank them by their semantic content
For example, you could feed each possible configuration through GNU Go, asking it to score the configurations, and pick the configuration which is most ‘intelligent’/likely to be produced by strong scorers. Or, you could instead tell GNU Go that ‘black is a 20 kyu player, white is 1 dan’ and rank configurations by which configuration is most consistent with such a differential (configurations with stupid moves by black being far more likely than stupid moves by white, and the converse).