Open jvc56 opened 3 years ago
It's certainly not a totally straightforward problem. Such an evaluation for every resigned position would likely be expensive. There were some "score per turn" heuristics that were considered but ultimately we decided to just use a simple turn based metric for now. If a satisfactory sandbagging heuristic were proposed that was performant, we would definitely consider it.
I suppose more generally speaking, there could be some quantity of "how winning/losing" a position is and a quantity "how sandbagging" it is to resign a particular game.