Data Leakage.

Before reading backtest returns, check for future data and missing costs

What could the model know at the time, and how were orders assumed to fill? Check time splits, preprocessing, slippage and funding before treating a return curve as evidence.