New Stacked Models Performances - Classification and Regression - Stock Time Series

Hello community.
Time ago i was looking for help on models to add to my new project.
Now i developed this new verison of stacked models for regression and classification.
After tweaking, tuning and training over 3200 assets with fixed date block for the train at 1/1/2023 to avid any possible timetravel here are the results.
We go in production next week. But i have a dounbt on one of them that is an LSTM with attention mechanis. This model that is in pic 3 predict the price move in regression with a time swich of 7 days so 7 days before we have the correct enforcement of the position. What is the best practice in this case ? We impose a date swich manually on the predictions ? Is amazing but 7 days before it acna also cause errors if not handled. Tnks also to all the people that helped me with very good tips on how to proceed at the beginning.

Report EvE - Report EvE

LSTM model with Time Shift - LSTM model time shift