Integrate and benchmark a differentiable delta-hyperbolicity measure within Graph Neural Networks, applying Riemannian geometry concepts to tackle over-squashing and boost geometric deep learning robustness and explainability.