Monitoring begins before every outcome is available. You can observe input and score distributions immediately, but fraud disputes or travel cancellations may mature later. You will measure an input shift, count completed label windows and evaluate probability loss only where the outcome is ready. A pending label isn't a negative event. Input drift can occur without performance deterioration, and performance can deteriorate without a large univariate shift. These diagnostics support an investigation rather than proving a cause. Keep a fixed reference population, a clear observation window and an explicit count of mature outcomes beside every report of model quality.