Overlapping five-day labels

Each daily row predicts the next five trading days' return. How would you prevent target overlap from contaminating a chronological validation split?

Associate every row with its information and label intervals. Before fitting, remove training rows whose label intervals overlap the validation interval or whose labels would not yet be known at the fitting date. Use a gap justified by actual horizons and feature availability, not an arbitrary fixed number. If training samples can follow a test block, assess an additional embargo. Fit preprocessing only on the retained training rows.