Breaking the Exploration Bottleneck: Rubric-Scaffolded Reinforcement Learning for General LLM Reasoning

TL;DR

This paper introduces RuscaRL, a novel reinforcement learning framework that uses rubric-based scaffolding to enhance exploration and reasoning in large language models, overcoming traditional limitations.

Contribution

RuscaRL employs explicit rubric-guided exploration and verifiable rewards to improve reasoning capabilities in LLMs, representing a new approach to RL training for general reasoning tasks.

Findings

RuscaRL outperforms baseline models on multiple reasoning benchmarks.

Rubric-guided exploration increases diversity and quality of responses.

Decaying guidance helps models internalize reasoning patterns.

Abstract

Recent advances in Large Language Models (LLMs) have underscored the potential of Reinforcement Learning (RL) to facilitate the emergence of reasoning capabilities. Despite the encouraging results, a fundamental dilemma persists as RL improvement relies on learning from high-quality samples, yet the exploration for such samples remains bounded by the inherent limitations of LLMs. This, in effect, creates an undesirable cycle in which what cannot be explored cannot be learned. In this work, we propose Rubric-Scaffolded Reinforcement Learning (RuscaRL), a novel instructional scaffolding framework designed to break the exploration bottleneck for general LLM reasoning. Specifically, RuscaRL introduces checklist-style rubrics as (1) explicit scaffolding for exploration during rollout generation, where different rubrics are provided as external guidance within task instructions to steer diverse high-quality responses. This guidance is gradually decayed over time, encouraging the model to internalize the underlying reasoning patterns; (2) verifiable rewards for exploitation during model training, where we can obtain robust LLM-as-a-Judge scores using rubrics as references, enabling effective RL on general reasoning tasks. Extensive experiments demonstrate the superiority of the proposed RuscaRL across various benchmarks, effectively expanding reasoning boundaries under the Best-of-N evaluation. Our code is available at https://github.com/IANNXANG/RuscaRL.