Gradient-based explanation methods are often perceived as the gold standard in machine learning interpretability.