frontier models
The Reasoning Step Loses Its Shortcut
CausalCoT isolates each atomic sub-question with only its direct parents, blocking attention paths to unrelated answer-bearing context.
Summary
CausalCoT isolates each atomic sub-question with only its direct parents, blocking attention paths to unrelated answer-bearing context.
A fluent chain of thought may explain an answer after the model has already recovered it directly from the full prompt. CausalCoT decomposes a problem into a directed acyclic graph and evaluates each node in a context containing only its causal parents. That removes attention paths from a local step to non-parent evidence. Across the reported reasoning suite, predictive performance remained competitive while the causal sensitivity of intermediate steps rose by 12 percentage points on average and as much as 21 points over the strongest baseline. Local corrections also propagated through downstream nodes. The method tests one notion of explanation faithfulness, not access to a model's complete internal cause.
Why it matters
CausalCoT isolates each atomic sub-question with only its direct parents, blocking attention paths to unrelated answer-bearing context.
Limits and context
- CausalCoT decomposes a problem into a directed acyclic graph and evaluates each node in a context containing only its causal parents.
- The method tests one notion of explanation faithfulness, not access to a model's complete internal cause.
Key claims
CausalCoT isolates each atomic sub-question with only its direct parents, blocking attention paths to unrelated answer-bearing context.
Qualification: CausalCoT decomposes a problem into a directed acyclic graph and evaluates each node in a context containing only its causal parents.
Evidence: source-2026-10-23-010
Sources
- arXiv preprint 2610.31639, original v1arXiv · primary research
Corrections
No corrections have been recorded for this story.