Scorers are a fundamental building block of many data valuation methods. They are typically used by ModelUtility and its subclasses to evaluate the quality of a model when trained on subsets of the training data.
Scorers evaluate trained models in user-defined ways, and provide additional information about themselves, like their range and default value, which can be used by some data valuation methods (e.g. Group Testing Shapley) to estimate the number of samples required for a certain quality of approximation.