interpretabilitycritiqueneutralExisting sufficiency-oriented explanation methods can assign high importance to spurious subsequences that support predictions without being essential to the model's decision.Machine Learning29 Jul 2026http://arxiv.org/abs/2607.21573v1