Diagnosing Multi-step Reasoning Failures in Black-box LLMs via Stepwise Confidence Attribution
This paper introduces Stepwise Confidence Attribution (SCA), a framework for closed-source LLMs that assigns step-level confidence based solely on generated reasoning traces. Using the Information Bottleneck principle, it identifies low-confidence steps correlated with errors, improving self-correction success rate by up to 13.5%.
[2605.19228] Diagnosing Multi-step Reasoning Failures in Black-box LLMs via Stepwise Confidence Attribution
[Submitted on 19 May 2026]
Title:Diagnosing Multi-step Reasoning Failures in Black-box LLMs via Stepwise Confidence Attribution
View a PDF of the paper titled Diagnosing Multi-step Reasoning Failures in Black-box LLMs via Stepwise Confidence Attribution, by Xiaoou Liu and 5 other authors
View PDF HTML (experimental)
Abstract:Large Language Models have achieved strong performance on reasoning tasks with objective answers by generating step-by-step solutions, but diagnosing where a multi-step reasoning trace might fail remains difficult. Confidence estimation offers a diagnostic signal, yet existing methods are restricted to final answers or require internal model access. In this paper, we introduce Stepwise Confidence Attribution (SCA), a framework for closed-source LLMs that assigns step-level confidence based only on generated reasoning traces. SCA applies the Information Bottleneck principle: steps aligning with consensus structures across correct solutions receive high confidence, while deviations are flagged as potentially erroneous. We propose two complementary methods: (1) NIBS, a non-parametric IB approach measuring consistency without graph structures, and (2) GIBS, a graph-based IB model that learns subgraphs through a differentiable mask to capture logical variability. Extensive experiments on mathematical reasoning and multi-hop question answering show that SCA reliably identifies low-confidence steps strongly correlated with reasoning errors. Moreover, using step-level confidence to guide self-correction improves the correction success rate by up to 13.5\% over answer-level feedback.
Comments: Accepted by ICML 2026
Subjects:
Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Information Theory (cs.IT); Machine Learning (cs.LG)
MSC classes: 68T50, 68T37, 68Q32
ACM classes: I.2.7; I.2.6; I.2.4
Cite as: arXiv:2605.19228 [cs.CL]
(or arXiv:2605.19228v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2605.19228
arXiv-issued DOI via DataCite (pending registration)
Submission history
From: Hua Wei [view email] [v1] Tue, 19 May 2026 00:57:51 UTC (520 KB)
Full-text links:
Access Paper:
View a PDF of the paper titled Diagnosing Multi-step Reasoning Failures in Black-box LLMs via Stepwise Confidence Attribution, by Xiaoou Liu and 5 other authors
View PDF
HTML (experimental)
TeX Source
view license
Current browse context:
cs.CL
new | recent | 2026-05
Change to browse by:
cs cs.AI cs.IT cs.LG math math.IT
References & Citations
NASA ADS
Google Scholar
Semantic Scholar
Loading...
Data provided by:
Bibliographic Tools
Bibliographic and Citation Tools
Bibliographic Explorer Toggle
Bibliographic Explorer (What is the Explorer?)
Connected Papers Toggle
Connected Papers (What is Connected Papers?)
Litmaps Toggle
Litmaps (What is Litmaps?)
scite.ai Toggle
scite Smart Citations (What are Smart Citations?)
Code, Data, Media
Code, Data and Media Associated with this Article
alphaXiv Toggle
alphaXiv (What is alphaXiv?)
Links to Code Toggle
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub Toggle
DagsHub (What is DagsHub?)
GotitPub Toggle
Gotit.pub (What is GotitPub?)
Huggingface Toggle
Hugging Face (What is Huggingface?)
ScienceCast Toggle
ScienceCast (What is ScienceCast?)
Demos
Demos
Replicate Toggle
Replicate (What is Replicate?)
Spaces Toggle
Hugging Face Spaces (What is Spaces?)
Spaces Toggle
TXYZ.AI (What is TXYZ.AI?)
Related Papers
Recommenders and Search Tools
Link to Influence Flower
Influence Flower (What are Influence Flowers?)
Core recommender toggle
CORE Recommender (What is CORE?)
Author
Venue
Institution
Topic
About arXivLabs
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?)