When evaluating a model, the sample period should cover multiple market states, including low-volatility periods, crisis periods, and sector rotation periods; validation only during bull markets cannot reflect real deployment performance