It sounds like you have a weekly seasonality in your data. The number of historical rows is calculated by adding the longest periodicity in your dataset to the feature derivation window. From the docs: "In this example, the prediction dataset needs at least 42 days of historical data and can predict (return) up to 7 rows. That is because although the model was configured for 35 days before the forecast point, seven days are added to the required history because the model uses seven-day differencing. Generally, Historical rows = FDW size + seasonality, where seasonality is the longest periodicity detected."