Label merchant emails and log categories before daily retrain

Your merchant emails arrive untagged, causing noisy training labels and delayed model updates. Data scientists receive clean labeled examples and a dated log to refresh models before the next daily retrain.