distribution gap

The difference between the training data distribution and the expected application domain, impacting a model's performance and generalization.