Hi , great work! I wanted to quickly ask. When augmenting lags and rolling features and then you split test and train, isn't there significant leakage in the test set due to the rolling features seeing the actual values for turnover?