How would you validate that a fraud model is not overfitting to historical dispute patterns at Chime?
