Tag
BACON proposes a four-stage pipeline that combines budgeted human labels with multiple AI judge outputs to produce calibrated item-level surrogate predictions, supporting both population-level estimation and individual-level scoring with improved accuracy and reduced bias.