Backtest

Walk-forward out-of-sample simulation. The model is trained on data ending before the test window, then predictions are evaluated on unseen data — no look-ahead bias.