---
format: "aidr-story-markdown/v1"
id: "f080970548be7b725c778bb6482a17450127b4f9151b0f4daf4df2ffd562a581"
canonical_url: "https://aidr.today/f0809705?lang=en"
title: "When Does External Guidance Help LLM Reasoning? A Bias-Variance Theory of Guidance-Augmented GRPO"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-10-07T04:00:00.000Z"
category: "Research"
topics: ["reasoning"]
source_urls: ["https://arxiv.org/abs/2610.06861"]
summary: "Researchers proposed Guidance-Augmented GRPO (GA-GRPO), a theoretical framework that casts external guidance as a stochastic operator rewriting the question distribution and analyzes the resulting policy-gradient estimator as a biased on-policy estimator whose bias is bounded by the total-variation guidance divergence. The framework subsumes vanilla GRPO, LUFFY, ExPO, PAPO, and TAPO as special cases."
---

# When Does External Guidance Help LLM Reasoning? A Bias\-Variance Theory of Guidance\-Augmented GRPO

> [Open the canonical story](<https://aidr.today/f0809705?lang=en>)

**Published:** 2026-10-07T04:00:00.000Z
**Category:** Research
**Topics:** reasoning

## Summary

Researchers proposed Guidance\-Augmented GRPO \(GA\-GRPO\), a theoretical framework that casts external guidance as a stochastic operator rewriting the question distribution and analyzes the resulting policy\-gradient estimator as a biased on\-policy estimator whose bias is bounded by the total\-variation guidance divergence\. The framework subsumes vanilla GRPO, LUFFY, ExPO, PAPO, and TAPO as special cases\.

## Sources

- [Story source](<https://arxiv.org/abs/2610.06861>)