Quantifying how confident a model is in its predictions, critical for safe deployment in high-stakes applications.