# Failure-Guided Co-Evolution of Prompts and Training Data

arXiv: https://arxiv.org/abs/2609.15209 · HTML: https://mlsanity.com/paper/2609.15209/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 22 of its 43 references that are in the map.
- First posted on arXiv 14 Sep 2026.

## Builds on (most cited first, 22 of 22)

- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering](https://mlsanity.com/paper/1809.09600/) (2018)
- [WizardLM: Empowering large pre-trained language models to follow complex instructions](https://mlsanity.com/paper/2304.12244/) (2023)
- [DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence](https://mlsanity.com/paper/2606.19348/) (2026)
- [Promptbreeder: Self-Referential Self-Improvement Via Prompt Evolution](https://mlsanity.com/paper/2309.16797/) (2023)
- [Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models](https://mlsanity.com/paper/2510.04618/) (2025)
- [Magpie: Alignment Data Synthesis from Scratch by Prompting Aligned LLMs with Nothing](https://mlsanity.com/paper/2406.08464/) (2024)
- [Absolute Zero: Reinforced Self-play Reasoning with Zero Data](https://mlsanity.com/paper/2505.03335/) (2025)
- [Meta-Harness: End-to-End Optimization of Model Harnesses](https://mlsanity.com/paper/2603.28052/) (2026)
- [Optimizing Instructions and Demonstrations for Multi-Stage Language Model Programs](https://mlsanity.com/paper/2406.11695/) (2024)
- [HoVer: A Dataset for Many-Hop Fact Extraction And Claim Verification](https://mlsanity.com/paper/2011.03088/) (2020)
- [FiNER: Financial Numeric Entity Recognition for XBRL Tagging](https://mlsanity.com/paper/2203.06482/) (2022)
- [Meta Context Engineering via Agentic Skill Evolution](https://mlsanity.com/paper/2601.21557/) (2026)
- [CodecLM: Aligning Language Models with Tailored Synthetic Data](https://mlsanity.com/paper/2404.05875/) (2024)
- [AutoPDL: Automatic Prompt Optimization for LLM Agents](https://mlsanity.com/paper/2504.04365/) (2025)
- [DataEnvGym: Data Generation Agents in Teacher Environments with Student Feedback](https://mlsanity.com/paper/2410.06215/) (2024)
- [Procedural Environment Generation for Tool-Use Agents](https://mlsanity.com/paper/2506.11045/) (2025)
- [AMPO: Automatic Multi-Branched Prompt Optimization](https://mlsanity.com/paper/2410.08696/) (2024)
- [DISCERN: Decoding Systematic Errors in Natural Language for Text Classifiers](https://mlsanity.com/paper/2410.22239/) (2024)
- [Forewarned is Forearmed: Leveraging LLMs for Data Synthesis through Failure-Inducing Exploration](https://mlsanity.com/paper/2410.16736/) (2024)
- [SIPDO: Closed-Loop Prompt Optimization via Synthetic Data Feedback](https://mlsanity.com/paper/2505.19514/) (2025)
- [Data-Prompt Co-Evolution: Growing Test Sets to Refine LLM Behavior](https://mlsanity.com/paper/2510.12728/) (2025)

## Abstract (arXiv)

Automatic prompt optimization (APO) improves language-model programs by revising prompts from task feedback, yet it typically holds its training data fixed. Repeatedly optimizing against the same instances confines feedback to weaknesses already represented in those data, leaving related failure conditions unexplored. We therefore view each failure as a dual signal: it indicates both how the prompt should be revised and what new training evidence should be synthesized. We introduce FORGE, a failure-guided framework that co-evolves prompts and training data. FORGE abstracts imperfect executions into reusable failure modes and synthesizes new training data through four complementary mutation strategies. Verified instances are fed back into prompt search, allowing updated prompts to expose the next data needs. Across eight heterogeneous benchmarks, FORGE improves the aggregate score over the unoptimized baseline by 16.52 percentage points and outperforms all evaluated APO baselines. The synthesized data also transfer beyond FORGE: in a transfer study, they improve all nine APO comparisons by 2--9 points and all three GRPO comparisons by 4--8 points under matched optimization budgets. These results establish failures as a shared interface between prompt optimization and data synthesis, and show the benefit of jointly adapting what a model is instructed to do and what it learns from.