![[2602.22583] Strategy Executability in Mathematical Reasoning: Leveraging Human-Model Differences for Effective Guidance](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.22583] Strategy Executability in Mathematical Reasoning: Leveraging Human-Model Differences for Effective Guidance
Summary
This paper explores the concept of strategy executability in mathematical reasoning, highlighting the differences between human and model strategies and introducing a framework for improved guidance.
Why It Matters
Understanding the gap between strategy usage and executability is crucial for enhancing AI models' performance in mathematical reasoning tasks. The proposed Selective Strategy Retrieval (SSR) framework offers a systematic approach to improve accuracy, making it relevant for researchers and practitioners in AI and education.
Key Takeaways
- Identifies the instability in example-based guidance for mathematical reasoning.
- Differentiates between strategy usage and executability in AI models.
- Introduces Selective Strategy Retrieval (SSR) to enhance model performance.
- Demonstrates SSR's effectiveness across multiple reasoning benchmarks.
- Provides publicly available code and benchmarks for further research.
Computer Science > Artificial Intelligence arXiv:2602.22583 (cs) [Submitted on 26 Feb 2026] Title:Strategy Executability in Mathematical Reasoning: Leveraging Human-Model Differences for Effective Guidance Authors:Weida Liang, Yiyou Sun, Shuyuan Nan, Chuang Li, Dawn Song, Kenji Kawaguchi View a PDF of the paper titled Strategy Executability in Mathematical Reasoning: Leveraging Human-Model Differences for Effective Guidance, by Weida Liang and 5 other authors View PDF HTML (experimental) Abstract:Example-based guidance is widely used to improve mathematical reasoning at inference time, yet its effectiveness is highly unstable across problems and models-even when the guidance is correct and problem-relevant. We show that this instability arises from a previously underexplored gap between strategy usage-whether a reasoning strategy appears in successful solutions-and strategy executability-whether the strategy remains effective when instantiated as guidance for a target model. Through a controlled analysis of paired human-written and model-generated solutions, we identify a systematic dissociation between usage and executability: human- and model-derived strategies differ in structured, domain-dependent ways, leading to complementary strengths and consistent source-dependent reversals under guidance. Building on this diagnosis, we propose Selective Strategy Retrieval (SSR), a test-time framework that explicitly models executability by selectively retrieving and combining str...
