The system measures active recommendations by querying the model with forced-choice recommendation questions across repeated trials to calculate an empirical relevance percentage score.