# SWE-Critix trains agent critics for code repair

> The method trains a code agent to critique repair trajectories using differential reasoning and test rewards.

- Canonical URL: https://newsverum.com/en/news/swe-critix-trains-agent-critics-on-code-repair-trajectories
- Publisher: Newsverum
- Section: Technology
- Country: Hong Kong SAR China
- Published: 2026-10-13T00:51:41+00:00
- Updated: 2026-10-13T00:51:41+00:00
- Event / data period: 2026-10-09
- Other editions: [Türkçe](https://newsverum.com/tr/news/swe-critix-trains-agent-critics-on-code-repair-trajectories.md), [العربية](https://newsverum.com/ar/news/swe-critix-trains-agent-critics-on-code-repair-trajectories.md), [Français](https://newsverum.com/fr/news/swe-critix-trains-agent-critics-on-code-repair-trajectories.md), [Español](https://newsverum.com/es/news/swe-critix-trains-agent-critics-on-code-repair-trajectories.md), [Deutsch](https://newsverum.com/de/news/swe-critix-trains-agent-critics-on-code-repair-trajectories.md), [Português](https://newsverum.com/pt/news/swe-critix-trains-agent-critics-on-code-repair-trajectories.md)

Researchers affiliated with Huawei and Monash University Malaysia presented SWE-Critix, a training method for verifiers that judge whether a coding agent’s execution trajectory completed its task. A teacher model compares successful and failed trajectories for the same issue to produce training explanations, followed by supervised fine-tuning and reinforcement learning with rewards based on test outcomes.

The final system reached 78% overall accuracy in the paper’s CodeForge Preview evaluation. The authors measured this on a test set derived from the same data source; they say test-reward training helped distinguish successful repairs, while recall remained low for some error classes. This early benchmark result does not establish repair success on real repositories.

## Data

### Research result

CodeForge Preview; test trajectories derived from the same source; overall accuracy reported by authors

Methodology: The value is limited to the benchmark experiment reported in the paper; it is not a field result or general performance guarantee.

| Measure | Percent | Scope |
|---|---|---|
| Overall accuracy on CodeForge Preview test trajectories | 78 | CodeForge Preview; test trajectories derived from the same source; overall accuracy reported by authors |

Sources: <https://arxiv.org/html/2610.13356v1> · Download data: [JSON](https://newsverum.com/api/v1/articles/swe-critix-trains-agent-critics-on-code-repair-trajectories/data/reported-result-1?lang=en&format=json) · [CSV](https://newsverum.com/api/v1/articles/swe-critix-trains-agent-critics-on-code-repair-trajectories/data/reported-result-1?lang=en&format=csv)

## Sources

- [arXiv preprint by researchers affiliated with Huawei and Monash University Malaysia](<https://arxiv.org/html/2610.13356v1>) (Source checked: 2026-10-13; Published: 2026-10-09)

---

Cite the canonical URL above. [Editorial standards and corrections](https://newsverum.com/en/editorial)