Not sure your AI scorer agrees with trained examiners? Send 20 learner responses it has already scored, plus your band descriptors. I mark them blind as a second rater, compare my scores with your model's, and send a one-page indicative verdict: how closely we agree on this sample, where the biggest gaps sit (by task, band and language feature), and the three fixes worth making first. It is a fast diagnostic, not a reliability study. Includes a 20-minute readout call. The fee is credited against an AI Scorer Drift Audit booked within 30 days.