Hindsight Hint Distillation: Scaffolded Reasoning for SWE Agents from CoT-free Answers

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wang, Shengjie, Li, Guanghe, Yang, Zonghan, Gao, Yang
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917483708940288
author Wang, Shengjie
Li, Guanghe
Yang, Zonghan
Gao, Yang
author_facet Wang, Shengjie
Li, Guanghe
Yang, Zonghan
Gao, Yang
contents Solving complex long-horizon tasks requires strong planning and reasoning capabilities. Although datasets with explicit chain-of-thought (CoT) rationales can substantially benefit learning, they are costly to obtain. To address this challenge, we propose Hindsight Hint Distillation (HHD), which only requires easy-to-obtain question-answer pairs without CoT annotations. Inspired by how human teachers use student mistakes to provide targeted guidance, HHD synthesizes hindsight hints from the model's own failed self-rollouts and uses them to scaffold on-policy rollouts that successfully complete the tasks. The model then self-distills these scaffolded trajectories and generalizes to new problems without hint guidance. Experiments show that HHD significantly outperforms iterative RFT and trajectory-synthesis baselines, achieving an absolute improvement of 8\% on SWE-bench Verified, while all baselines improve by only around 2\%. Notably, the reasoning strategies induced by HHD generalize effectively to out-of-distribution tasks, yielding the largest gains on SWE-bench Multilingual despite no training on multilingual data. These results demonstrate that HHD can effectively synthesize expert-like reasoning from CoT-free data and substantially improve long-horizon performance.
format Preprint
id arxiv_https___arxiv_org_abs_2605_11556
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Hindsight Hint Distillation: Scaffolded Reasoning for SWE Agents from CoT-free Answers
Wang, Shengjie
Li, Guanghe
Yang, Zonghan
Gao, Yang
Artificial Intelligence
Machine Learning
Solving complex long-horizon tasks requires strong planning and reasoning capabilities. Although datasets with explicit chain-of-thought (CoT) rationales can substantially benefit learning, they are costly to obtain. To address this challenge, we propose Hindsight Hint Distillation (HHD), which only requires easy-to-obtain question-answer pairs without CoT annotations. Inspired by how human teachers use student mistakes to provide targeted guidance, HHD synthesizes hindsight hints from the model's own failed self-rollouts and uses them to scaffold on-policy rollouts that successfully complete the tasks. The model then self-distills these scaffolded trajectories and generalizes to new problems without hint guidance. Experiments show that HHD significantly outperforms iterative RFT and trajectory-synthesis baselines, achieving an absolute improvement of 8\% on SWE-bench Verified, while all baselines improve by only around 2\%. Notably, the reasoning strategies induced by HHD generalize effectively to out-of-distribution tasks, yielding the largest gains on SWE-bench Multilingual despite no training on multilingual data. These results demonstrate that HHD can effectively synthesize expert-like reasoning from CoT-free data and substantially improve long-horizon performance.
title Hindsight Hint Distillation: Scaffolded Reasoning for SWE Agents from CoT-free Answers
topic Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2605.11556