A model's future voltage depends on both its current voltage and a slowly decaying adaptation variable. Why is tracking voltage alone insufficient to predict its future behavior?