# When Models Edit Too Much: On the Fidelity of Minimal Code Edits

arXiv: https://arxiv.org/abs/2609.04061 · HTML: https://mlsanity.com/paper/2609.04061/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Builds on 22 of its 66 references that are in the map.
- First posted on arXiv 3 Sep 2026.

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

- [Evaluating Large Language Models Trained on Code](https://mlsanity.com/paper/2107.03374/) (2021)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Program Synthesis with Large Language Models](https://mlsanity.com/paper/2108.07732/) (2021)
- [SWE-bench: Can Language Models Resolve Real-World GitHub Issues?](https://mlsanity.com/paper/2310.06770/) (2023)
- [LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code](https://mlsanity.com/paper/2403.07974/) (2024)
- [LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Models](https://mlsanity.com/paper/2403.13372/) (2024)
- [Is Your Code Generated by ChatGPT Really Correct? Rigorous Evaluation of Large Language Models for Code Generation](https://mlsanity.com/paper/2305.01210/) (2023)
- [SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training](https://mlsanity.com/paper/2501.17161/) (2025)
- [CodeBLEU: a Method for Automatic Evaluation of Code Synthesis](https://mlsanity.com/paper/2009.10297/) (2020)
- [TACO: Topics in Algorithmic COde generation dataset](https://mlsanity.com/paper/2312.14852/) (2023)
- [Benchmarking Complex Instruction-Following with Multiple Constraints Composition](https://mlsanity.com/paper/2407.03978/) (2024)
- [Out of the BLEU: how should we assess quality of the Code Generation models?](https://mlsanity.com/paper/2208.03133/) (2022)
- [When Thinking Fails: The Pitfalls of Reasoning for Instruction-Following in LLMs](https://mlsanity.com/paper/2505.11423/) (2025)
- [Can It Edit? Evaluating the Ability of Large Language Models to Follow Code Editing Instructions](https://mlsanity.com/paper/2312.12450/) (2023)
- [SWE-Perf: Can Language Models Optimize Code Performance on Real-World Repositories?](https://mlsanity.com/paper/2507.12415/) (2025)
- [HumanEvo: An Evolution-aware Benchmark for More Realistic Evaluation of Repository-level Code Generation](https://mlsanity.com/paper/2406.06918/) (2024)
- [CREF: An LLM-based Conversational Software Repair Framework for Programming Tutors](https://mlsanity.com/paper/2406.13972/) (2024)
- [EDIT-Bench: Evaluating LLM Abilities to Perform Real-World Instructed Code Edits](https://mlsanity.com/paper/2511.04486/) (2025)
- [CoreCodeBench: Decoupling Code Intelligence via Fine-Grained Repository-Level Tasks](https://mlsanity.com/paper/2507.05281/) (2025)
- [PAFT: Preservation Aware Fine-Tuning for Minimal-Edit Program Repair](https://mlsanity.com/paper/2604.03113/) (2026)
- [QiMeng-PRepair: Precise Code Repair via Edit-Aware Reward Optimization](https://mlsanity.com/paper/2604.05963/) (2026)
- [Edit, But Verify: An Empirical Audit of Instructed Code-Editing Benchmarks](https://mlsanity.com/paper/2604.05100/) (2026)

## Built on by (most cited first, 3 of 3)

- [ParanoiaEval: Benchmarking Unnecessary Defensive Work in Agentic Coding](https://mlsanity.com/paper/2610.08662/) (2026)
- [SWE-MISBEHAVIOR: Benchmarking and Mitigating Boundary Violations in Software Coding Agents](https://mlsanity.com/paper/2610.18206/) (2026)
- [PreciseCoder: Faithful and Generalizable Debugging Beyond Unit-Test Rewards](https://mlsanity.com/paper/2610.18755/) (2026)

## Abstract (arXiv)

Large language models (LLMs) are increasingly used to edit existing code, but correctness alone is not enough: useful repairs should also be minimal, reviewable, and faithful to the original implementation. We study over-editing, the tendency of a model to rewrite code beyond what is required to fix a bug. We construct an evaluation framework from 400 BigCodeBench problems by injecting controlled AST-level corruptions into reference solutions, giving each repair task a known minimal patch. Across frontier LLMs, over-editing is widespread even among strong models like GPT-5.5: high Pass@1 can coexist with unnecessarily large edits and added cognitive complexity. A preservation instruction substantially reduces this behavior, lowering average excess Levenshtein distance from 0.195 to 0.131, reducing added cognitive complexity by 26.6%, and increasing Pass@1 by 2.3 points. However, these gains do not simply follow from a larger reasoning budget or larger models. We next ask whether minimal editing can be learned directly during post-training. We observe that supervised fine-tuning overfits to seen corruption patterns, whereas reinforcement learning gives the best out-of-domain edit-fidelity and performance-retention trade-off. These results position edit fidelity as a distinct axis of code-repair quality and show that it can be measured and learned.