Circuit Insights: Towards Interpretability Beyond Activations