Right now, when a model is built, several variations of the model can be built in order to select the hyperparameters that result in the best model according to some evaluation metric. However it might also be useful to further require that the metric meet some threshold before it's published. That way, if the 'best' model were quite bad, it would be rejected.