@article{whymultisteptoolusereinforcementlearning,
  title = {Why Multi-Step Tool-Use Reinforcement Learning Collapses and How Supervisory Signals Fix It},
  author = {Yupu Hao and Zhuoran Jin and Huanxuan Liao and Kang Liu and Jun Zhao},
  year = {2026},
  eprint = {2606.26027},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2606.26027},
}