The article tests a new method to check if a model accurately predicts outcomes in real-world decisions. Instead of just looking at how well the model fits the data, this method also considers how well the decisions based on the model actually perform. The researchers applied this method to pricing consumer loans and found that traditional tests may reject simple models, like the logit model, even though they can still make good decisions. This shows that sometimes, even if a model is not perfect, it can still lead to good results in practice.