@misc{indiciae59e33dbccca9,
  title = {Erased, Rerouted, or Rescaled? Post-Training and the Causal Quotient of a Language Model's Belief State},
  author = {Weihan Li and Tianshi Zheng and Junhao Wu and Xinlei Chen},
  year = {2026},
  url = {https://arxiv.org/abs/2610.05292},
  note = {Source identifier: 2610.05292}
}