cs.CLNov 3, 2026

ReSI-Math: Exploring Recoverable Capability Frontiers for Efficient Reasoning Improvement

Authors: Yao Cui, Zhen Yang, Yuandong Wang, Yangfu Zhu, Tianyu Wo, Zhenzhou Shao

Organizations: Capital Normal University · Tsinghua University · Beihang University

Abstract

Reinforcement learning with verifiable rewards (RLVR) has become an effective approach for improving mathematical reasoning in large language models. However, a fundamental question remains: when a model fails to solve a problem, does it truly lack the underlying capability? In this work, we reveal that direct failure does not necessarily imply inability. Some reasoning failures can be recovered by exploring the model's own trajectories, exposing a recoverable capability frontier: a policy-dependent region where solutions are not reliably generated through direct decoding but can be elicited through trajectory-level search. Based on this observation, we introduce ReSI-Math, a framework that explores recoverable capability frontiers and converts recovered capabilities into data-efficient RLVR supervision. ReSI-Math uses Self-Judge-guided evolutionary search to identify recoverable failures under a fixed policy and adaptively selects frontier problems for sequential RLVR as the policy evolves. On Qwen3-VL-8B, correct-incorrect trajectory crossover produces 272 verifier-confirmed correct offspring from 621 attempts (43.8%). After three sequential one-shot RLVR stages, the MATH500 and WE-MATH average improves from 78.71 to 80.16. Under matched single-problem RLVR, random selection reaches only 53.31 versus 54.16 for frontier-guided Q1.

Figures & tables

Appendix figures & tables4 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

CardsList
  1. Right Makes Might: Aligning Verified Hidden States Empowers RL Reasoning

    Jun 2, 2026Ziyue Wang, Aomufei Yuan, Yongfu Zhu +10LLM AlignmentRL for Language Model Reasoning

  2. Improving Math Reasoning through Value-guided Informative Search

    Oct 1, 2026Shaohuai Liu, Yuning Wu, Haoran Liu +3RL for Language Model ReasoningReinforcement Learning with Verifiable Rewards