Code: <a href=\"https://github.com/nreHieW/over-editing\" rel=\"nofollow\">https://github.com/nreHieW/over-editing</a></p>\n","updatedAt":"2026-09-07T02:09:35.611Z","author":{"_id":"647db42711084fb58318b053","avatarUrl":"/avatars/901a4db8e99f6e0c7e7e596c251b315c.svg","fullname":"Tongyao","name":"tyzhu","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7536159157752991},"editors":["tyzhu"],"editorAvatarUrls":["/avatars/901a4db8e99f6e0c7e7e596c251b315c.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2609.04061","authors":[{"_id":"6a9a685d8f7c3b75572396aa","name":"Tongyao Zhu","hidden":false},{"_id":"6a9a685d8f7c3b75572396ab","name":"Wei Hern Lim","hidden":false},{"_id":"6a9a685d8f7c3b75572396ac","name":"Min-Yen Kan","hidden":false}],"publishedAt":"2026-09-03T00:00:00.000Z","submittedOnDailyAt":"2026-09-07T00:00:00.000Z","title":"When Models Edit Too Much: On the Fidelity of Minimal Code Edits","submittedOnDailyBy":{"_id":"647db42711084fb58318b053","avatarUrl":"/avatars/901a4db8e99f6e0c7e7e596c251b315c.svg","isPro":true,"fullname":"Tongyao","user":"tyzhu","type":"user","name":"tyzhu"},"summary":"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.","upvotes":2,"discussionId":"6a9a685e8f7c3b75572396ad","githubRepo":"https://github.com/nreHieW/over-editing","githubRepoAddedBy":"user","ai_summary":"Large language models frequently over-edit code during repair, but preservation instructions and reinforcement learning can improve edit fidelity without sacrificing correctness.","ai_keywords":["AST-level corruptions","over-editing","Levenshtein distance","cognitive complexity","preservation instruction","supervised fine-tuning","reinforcement learning","edit fidelity"],"ai_summary_model":"thinkingmachines/Inkling-Small","githubStars":19,"organization":{"_id":"6508ab2b349930913196378b","name":"NationalUniversityofSingapore","fullname":"National University of Singapore","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/630ca0817dacb93b33506ce7/ZYUmpSMsa5Whihw3me2Bw.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"647db42711084fb58318b053","avatarUrl":"/avatars/901a4db8e99f6e0c7e7e596c251b315c.svg","isPro":true,"fullname":"Tongyao","user":"tyzhu","type":"user"},{"_id":"63ac5701c21e60a3e9b58aa7","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63ac5701c21e60a3e9b58aa7/g6EX7diOpuA94R2ab-rZC.png","isPro":true,"fullname":"Dipankar Sarkar","user":"dipankarsarkar","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"6508ab2b349930913196378b","name":"NationalUniversityofSingapore","fullname":"National University of Singapore","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/630ca0817dacb93b33506ce7/ZYUmpSMsa5Whihw3me2Bw.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2609/2609.04061.md","query":{}}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2609.04061 in a model README.md to link it from this page.

Cite arxiv.org/abs/2609.04061 in a dataset README.md to link it from this page.

Cite arxiv.org/abs/2609.04061 in a Space README.md to link it from this page.

Add this paper to a collection to link it from this page.