TY  - RPRT
TI  - When Policies Cannot Be Retrained: A Unified Closed-Form View of Post-Training Steering in Offline Reinforcement Learning
AU  - Elias Hossain
AU  - Mohammad Jahid Ibna Basher
AU  - Ivan Garibay
AU  - Ozlem Garibay
AU  - Niloofar Yousefi
PY  - 2026
UR  - https://arxiv.org/abs/2604.22873
ID  - 2604.22873
ER  -