Thanks for bringing this up. I had a similar question when looking through the example. From what I understand, it can seem forward-looking at first because of how the data is prepared and the model is trained. The important part is whether the prediction at each point is based only on information that would have been available at that time. If that's the case, then it may not be true look-ahead bias, just a backtesting implementation detail. That said, it's definitely worth double-checking because data leakage in machine learning strategies can be surprisingly easy to miss and can make results look much better than they would be in live trading.
