Towards Long-Horizon Interpretability: Efficient and Faithful Multi-Token Attribution for Reasoning LLMs

Wenbo Pan, Zhichao Liu, Xianlong Wang, Yu Haining, Xiaohua Jia

SAI review preview

SAI reviewed this ICML 2026 paper and its available research artifacts. Sign in to read the complete referee report, inline feedback, and execution-based verification where available.

36Detailed review comments

CompletePaper and code review

19% of graded claims reproducedReplication status