@misc{indiciae07806db615f8,
  title = {What Will Post-Training Fix? Per-Problem Gains Are Shared Across Independent RL Runs, and Existing Checkpoints Predict Them Better Than A Priori Signals},
  author = {Xiaoxian Duan},
  year = {2026},
  url = {https://arxiv.org/abs/2610.04978},
  note = {Source identifier: 2610.04978}
}