CoChat

A Closed-Loop Modular Language Agent with Step Verification and Local Correction for Multi-Step Task Solving

Open access
Li, He, Feng, Lihang, Wang, Dong · 2026 · Electronics

The DOI matches its Crossref record; 39 of the 39 references it cites were found in a registry or an index; 21 of the 33 sources its claims cite were checked for support.

Abstract

Multi-step task solving with large language models in intelligent electronic systems and interactive environments requires stronger process control and execution reliability. To address local error accumulation during multi-step execution, this paper proposes a closed-loop modular language agent framework integrating task planning, action execution, step verification, local regeneration, and replanning. A process-supervision data construction method is further introduced, in which real execution steps are retained as valid samples and invalid samples are automatically synthesized through action substitution, input perturbation, observation replacement, and subgoal mismatch, providing step-level supervision for validity prediction. In the proposed framework, step verification functions as a process-level control signal that supports hierarchical recovery through local regeneration and replanning, rather than as a standalone filtering module. Experiments are conducted on mathematical reasoning and web interaction tasks. On mathematical reasoning tasks, the proposed framework achieves an accuracy of 0.650, compared with 0.317 for Integrated Training, 0.460 for CoT Training, 0.568 for ReFT, and 0.617 for Agent Lumos. On web interaction tasks, the proposed framework achieves a step success rate of 0.424, compared with 0.246 for Integrated Training and 0.310 for Agent Lumos. Among the cases where recovery is triggered, local regrounding succeeds in 85.7% of reground attempts, while replanning succeeds in 53.3% of replanning attempts. These results indicate that the proposed framework improves process stability and recovery capability in multi-step task solving.

Overview

CoChat Summary

## TL;DR

The paper proposes CL-SPF, a closed-loop modular language agent that adds step verification and hierarchical recovery to multi-step task solving so invalid intermediate steps are corrected before they propagate. Experiments on mathematical reasoning and web interaction show improved performance over baseline methods, especially when verification is coupled with local regrounding and replanning.

## Key findings

  • On mathematical reasoning tasks, CL-SPF achieved 0.650 accuracy, compared with 0.317 for Integrated Training, 0.460 for CoT Training, 0.568 for ReFT, and 0.617 for Agent Lumos.
  • On web interaction tasks, CL-SPF achieved a 0.424 step success rate, compared with 0.246 for Integrated Training and 0.310 for Agent Lumos; the paper also reports 0.315 and 0.324 for Agent Lumos with fixed retries of R = 6 and R = 7.
  • In the module ablation study on GSM, the full model achieved the highest Answer Accuracy (0.650); removing the verifier dropped accuracy to 0.567, removing regrounding to 0.608, and removing replanning to 0.633.
  • For recovery, the paper reports 85.7% reground success and 53.3% replan success among triggered recovery attempts; on GSM, 55.0% of cases were solved after recovery was triggered, 32.5% remained unresolved after correction, and 10.0% of all samples were correct without triggering correction.

## Methods

  • The framework is a closed-loop modular agent with four main modules: task planning, action execution, step verification, and recovery (local regeneration/regrounding, then replanning if needed).
  • Experiments were run on mathematical reasoning datasets (GSM, ASDiv, PRM800K) and the web interaction dataset Mind2Web using its Test-Domain split.
  • Step verification is trained as a binary classification problem using positive real execution steps and synthetic negative samples created by action substitution, input perturbation, observation replacement, and subgoal mismatch; the paper uses metrics at the task, step, and process levels.

## Limitations

  • The paper states that the study is limited to two representative task settings and a fixed base model; future work will extend to broader domains and stronger models.
  • It notes that verification and recovery add computational overhead through extra action attempts and replanning, and that direct wallclock latency and token consumption were not measured.
  • It also states that the verifier is trained on automatically constructed negative samples, which may not fully cover real-world failure diversity.
  • Evident from the text: several reported gains come with more action attempts / retries, so improved robustness is partly traded for extra execution cost.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3390/electronics15102011

Cites

39 of 39 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

13 Valid · 0 Not valid · 9 Not checked · 136 Own results and interpretations

60 more claims

Create a free account to see all 158 claims and to verify the 9 references still open.

See all claims — free account
  • Probably not a substantial claim. 37
  • It may state two separate claims as one. 62
  • The quote only partly states this claim. 3

Create a free account to see these claims.

Citations

Cites

Show all 39 references

Research works printed but not linked: 2

  1. Yao, S., Zhao, J., Yu, D., Du, N., Shafran, I., Narasimhan, K.R., and Cao, Y. (2023, January 1–5). ReAct: Synergizing Reasoning and Acting in Language Models. Proceedings of the Eleventh International Conference on Learning Representations (ICLR 2023), Kigali, Rwanda.
  2. Qin, Y., Liang, S., Ye, Y., Zhu, K., Yan, L., Lu, Y., Lin, Y., Cong, X., Tang, X., and Qian, B. (2024, January 7–11). ToolLLM: Facilitating Large Language Models to Master 16000+ Real-World APIs. Proceedings of the Twelfth International Conference on Learning Representations (ICLR 2024), Vienna, Austria.

Metadata

Authors
Li, H., Feng, L., & Wang, D.
Affiliations
  • School of Instrument Science and Engineering, Southeast University, Nanjing 210096, China
  • Nanjing Tech University (CN)
  • Southeast University (CN)
Corresponding author
Wang, Dong
Venue
Electronics
Publisher
MDPI AG
Published
2026-05-09
Volume
15
Issue
10
Pages
2011
ISSN
2079-9292 (ISSN-L)
OpenAlex
W7160826567
Cited by
0
References
13
FWCI
0.00
Type
journal article
Language
en
Open access
Locations
Electronics (publishedVersion, cc-by)
Funders
  • National Natural Science Foundation of China (62103184)
  • Jiangsu Science and Technology Programme (BZ2024057)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › AI-based Problem Solving and Planning
  • Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
  • Physical Sciences › Computer Science › Artificial Intelligence › Speech and dialogue systems
Keywords
large language models, local regeneration, replanning, process supervision, mathematical reasoning, web interaction

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol and qwen3.8-omni-flash).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: A Closed-Loop Modular Language Agent with Step Verification and Local Correction for Multi-Step Task Solving. https://app.cochat.ai/papers/10.3390/electronics15102011
The paper
Li, H., Feng, L., & Wang, D. (2026). A Closed-Loop Modular Language Agent with Step Verification and Local Correction for Multi-Step Task Solving. Electronics. https://doi.org/10.3390/electronics15102011

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.