@misc{indiciae2f683f7f8b7f,
  title = {Post-Training with Policy Gradients: Optimality and the Base Model Barrier},
  author = {Alireza Mousavi-Hosseini and Murat A. Erdogdu},
  year = {2026},
  url = {https://arxiv.org/abs/2603.06957},
  note = {Source identifier: 2603.06957}
}