Major concerns: 1 To R2&R4 on end-to-end GNN policy learning, and robustness of our two-stage framework w.r.t. detection