@article{rethinkingthecomparisonunitinsequencelev,
  title = {Rethinking the Comparison Unit in Sequence-Level Reinforcement Learning: An Equal-Length Paired Training Framework from Loss Correction to Sample Construction},
  author = {Fei Ding and Yongkang Zhang and Runhao Liu and Yuhao Liao and Zijian Zeng and Huiming Yang and Sibo wang and Linglin Liao},
  year = {2026},
  eprint = {2604.17328},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2604.17328},
}