1 paper
Chaoqun Cui, Hao Zhou, Meiqi Chen +2
Long-horizon software engineering (SWE) agents trained with reinforcement learning with verifiable rewards (RLVR) typically receive only terminal outcome supervision, making it dif…