By utilizing the information encoded in the connectivity of the nodes affected by an attack (idle network), the assessment of network robustness is systematically enhanced. Moreover, the idle network information provides generalizability to train models to estimate network robustness, even for previously unseen attacks.