Briar Retail needs a review recommendation for fictional address-correction cases. Two candidate predictions and a review_limit are supplied. Deliver prevalence, model A accuracy, model B recall and precision, false negatives, review count and the model selected by the stated coverage/capacity rule.
Briar Retail reviews orders that may need a manual address correction. Class 1 identifies an order with a confirmed correction. The fictional holdout contains many ordinary orders, so a model that raises no alerts can still appear accurate. The operations lead wants to find the correction cases while keeping the number of reviews within review_limit.
Two candidate label arrays are already supplied. Evaluate them without refitting or changing the holdout labels. You will calculate the positive rate, model A accuracy, model B recall and precision, then select the metric that matches the lead's stated priority. Finish with a checked recommendation based on recall and capacity. The selection rule makes the tradeoff explicit. Another team can choose a different priority, but its metric and capacity rule must describe that choice.