Test Cases for Impact Regularisation Methods — AI Alignment Forum