A prediction becomes useful when it changes a decision under explicit costs and constraints. Equal error costs justify one threshold; expensive missed events can justify another if probabilities are calibrated for the relevant population. You will calculate an expected-cost threshold, evaluate realized costs and build a queue with a fixed capacity. A capacity limit can override a threshold because the team can't process unlimited reviews. Ties need a repeatable rule. Keep the units and assumptions visible: the costs in these exercises are fictional decision inputs, not prices or measured business losses. A good ranking alone can't choose the policy for the operator.