When Reasoning Traces Conflict with Evidence: How Answer Configuration Shapes Evidence Following
Quick Answer
The study reveals that using a separate answer call improves evidence-following by 29.9 to 60.6 points across models like GPT-OSS, Gemma4, and Qwen3 on HotpotQA and MuSiQue.
Quick Take
While reasoning traces can enhance accuracy, they often conflict with current evidence, indicating that trace deletion alone cannot determine the source of an answer. Evaluations should report both properties alongside the answer configuration.
Key Points
- Separate answer calls outperform continuation in evidence-following by 29.9 to 60.6 points.
- Models tested include GPT-OSS, Gemma4, and Qwen3 on HotpotQA and MuSiQue.
- Trace deletion alone does not clarify which evidence an answer follows.
- Final line of a trace holds most value; gold evidence aids before trace creation.
- Renewed reasoning aligns with current evidence more effectively than verification instructions.
Paper Resources
📖 Reader Mode
~2 min readAbstract:A reasoning trace retained at answer time plays two roles: it can raise accuracy, and it can compete with the current evidence when the two disagree. We disentangle these roles by crossing verified evidence versions with the traces generated from each and answering from the same traces and evidence under two configurations: continuation of the model's own thinking and a separate answer call. Across three model families (gpt-oss, Gemma4, Qwen3) on HotpotQA and MuSiQue, the separate call raises evidence-following over continuation by 29.9 to 60.6 points. On the same pairs, removing the trace lowers accuracy by similar amounts under both configurations. Yet continuation repeats the stale conclusion in most conflicts, and the separate call follows the current evidence far more often. Benefiting from a trace and deferring to it are therefore distinct properties. Further analyses show that the final line of a trace carries most of its value, gold evidence helps most before the trace is written, distractors weaken correction under a separate call, and renewed reasoning, unlike a verification instruction, follows the current evidence in most conflicts. Because trace deletion alone cannot reveal which source an answer follows, evaluations should report both properties together with the answer configuration.
| Comments: | 25 pages, 4 figures |
| Subjects: | Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.20805 [cs.AI] |
| (or arXiv:2610.20805v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.20805 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Yanhang Shi [view email]
[v1]
Wed, 14 Oct 2026 18:52:23 UTC (232 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.AI
See more →VEPA: Verified Evolving Formal Policy for Lifelong Agent Guardrails
VEPA introduces a framework for verified evolving formal policies for LLM agents, enhancing safety by adapting guardrails based on deployment feedback. It outperforms existing methods across benchmarks like ASSEBench-Safety and ASSEBench-Security, ensuring deterministic policy enforcement even under challenging conditions.