@misc{indiciae97564b129f57,
  title = {No Single Tokenizer Feature Reliably Predicts Downstream Language Model Performance},
  author = {Antonio Lopardo and Luis Frentzen Salim and Shikhar Srivastava and Kevin Jiang and Mridul Sharma and Catherine Arnett},
  year = {2026},
  url = {https://arxiv.org/abs/2610.21679},
  note = {Source identifier: 2610.21679}
}