Systematic errors or tendencies in a model's predictions that are consistent across multiple samples for the same input.