Reflective Recovery: A Self-Supervised Method for Reasoning by Learning from Mistakes
Qirui Chen, Renjie Pi, Jiahui Gao and Lingpeng Kong (Zhejiang University, the University of Hong Kong and HKUST) convert failed reasoning rollouts into recovery training data, so a model learns to continue correctly from an already-wrong intermediate state.
Ask this paper
Scaling collapse is the target. Imitation learning on perfect trajectories stops improving once the problem set is fixed, because added positive examples teach nothing about states the model actually reaches at inference.
Method. Initial segments of failed trajectories are cut and concatenated with the prompt, then the model is guided to a valid solution from that contaminated prefix, producing supervision for error recognition without an external critic or reward model.
Results. On DeepSeek-R1-Distill-Qwen-7B accuracy rises from 30.0% to 37.5% on AIME 2025 and from 37.6% to 47.8% on Minerva.
Self-correction appears without being specified. The analyses report emergent correction behavior and continued improvement past the point where positive-only scaling flattens.
Significance. The training signal comes entirely from the model's own failures, which makes the method applicable wherever verified answers exist but step-level annotation does not.
Abstract
Data-driven fine-tuning is widely adopted to enhance reasoning in Large Language Models (LLMs) due to its simplicity and efficiency. However, mainstream imitation learning methods that rely exclusively on perfect reasoning trajectories suffer from a Scaling Collapse: when the problem set is limited, increasing positive examples fails to yield continuous improvement. However, during inference, an LLM can not guarantee that every intermediate step is correct and is therefore prone to errors. Once such errors arise, the LLM often struggles to recover and may be further misled by the accumulation of previous mistakes. To address this, we propose Reflective Recovery, a simple yet effective self-supervised approach that transforms failed reasoning attempts into recovery training data. Specifically, we extract initial segments of failed trajectories, concatenate them with prompts, and use them to guide the LLM toward valid solutions. Because these segments from failed trajectories are likely to contain errors, this process teaches models to recognize and correct mistakes during reasoning, enabling recovery from erroneous states without relying on external critics or reward models. Evaluated on extensive benchmarks, Reflective Recovery significantly improves performance. On DeepSeek-R1-Distill-Qwen-7B, it boosts accuracy from 30.0% to 37.5% on AIME 2025 and from 37.6% to 47.8% on Minerva. More importantly, analyses demonstrate that it breaks the scaling collapse barrier and enables models to develop emergent self-correction behaviors, representing a paradigm shift from outcome-oriented memorization to process-oriented reflective reasoning.