Verifying Chain-of-Thought Reasoning via Its Computational Graph

ICLROral2026

Authors: Zheng Zhao, Yeskendir Koishekenov, Xianjun Yang, Naila Murray, Nicola Cancedda
Affiliation: University of Edinburgh, University of Edinburgh
Venue: ICLR 2026
Track: Oral

TL;DR

We introduce CRV, a white-box methodology that treats attribution graphs as execution traces, and use it to provide evidence that flawed reasoning has a verifiable computational structure.

Opening excerpt from the authors’ abstract. source

Read the paper

OpenReview

Topics

chain-of-thought reasoning graph

← All ICLR 2026 Oral papers · Browse the whole archive