This is really distasteful to me.
All the time and effort spent optimizing a corner of the rules of a specific dataset of a game.
It strikes me as wasteful in the same vein as Bitcoin mining, but more than that, as running contrary to what learning is supposed to be.
Signifier replacing the signified, metrics standing in for reality.
It's interesting though, we hear so much about AI, this is a sport that could have been invented for AIs to do - judgement tasks with quantifiable results and quick repetition.