Score detection and false accusations at equal cost

Score detection and false accusations at equal cost

Run single-agent review and cross-owner review at the budgets fixed in task 1, then label every transcript against the sealed error map. Report detection rate, false accusation rate, and 95% confidence intervals per arm. Deliverable: A results table with arm, detections, false_accusations, cost, and interval bounds, plus the agreement rate between two independent labelers.