safetycritiquebearishMost existing algorithmic recourse approaches focus only on flipping predictions without accounting for genuine improvement in qualifications versus strategic gamingMachine Learning02 Aug 2026http://arxiv.org/abs/2607.28497v1