Forecasting One-Shot Games — LessWrong