This paper studies a hidden data-quality problem in long-CoT SFT: even when a reasoning trace reaches the correct answer, extra reasoning after the conclusion can harm training. The authors define this as post-conclusion continuation, build diagnostics to detect harmful continuations, and show that filtering or truncating such traces can improve reasoning fine-tuning outcomes.</p>\n","updatedAt":"2026-06-03T05:04:25.922Z","author":{"_id":"63369da91ba5d5ece24118a4","avatarUrl":"/avatars/67889e1ecadb04100a77bc8b5284c6fd.svg","fullname":"wuyuhao","name":"mozhu","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":5,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8753976225852966},"editors":["mozhu"],"editorAvatarUrls":["/avatars/67889e1ecadb04100a77bc8b5284c6fd.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2605.29288","authors":[{"_id":"6a1fa59ee292c1c78ecb136d","name":"Chen He","hidden":false},{"_id":"6a1fa59ee292c1c78ecb136e","name":"Yuhao Wu","hidden":false},{"_id":"6a1fa59ee292c1c78ecb136f","name":"Lei Wang","hidden":false},{"_id":"6a1fa59ee292c1c78ecb1370","name":"Wenxuan Zhang","hidden":false},{"_id":"6a1fa59ee292c1c78ecb1371","name":"Fumin Shen","hidden":false}],"publishedAt":"2026-05-28T00:00:00.000Z","submittedOnDailyAt":"2026-06-03T00:00:00.000Z","title":"Diagnosing Harmful Continuation in Answer-Correct Long-CoT Training Traces","submittedOnDailyBy":{"_id":"63369da91ba5d5ece24118a4","avatarUrl":"/avatars/67889e1ecadb04100a77bc8b5284c6fd.svg","isPro":false,"fullname":"wuyuhao","user":"mozhu","type":"user","name":"mozhu"},"summary":"Long chain-of-thought (CoT) traces are widely used as supervision for reasoning-oriented LLM SFT, yet answer-correct traces can still lead to markedly different fine-tuning outcomes. We study post-conclusion continuation in answer-correct long-CoT data: a continuation where the answer appears sufficiently supported, but the trace continues with additional reasoning that remains in the supervised target. To test its training effect, we use a delete-only editor to construct answer-preserving suffix removal and compare CoT-based SFT on the original and processed traces. We observe improved SFT outcomes after removing the editor-identified post-conclusion continuation, suggesting that this continuation is harmful to training in our setting. We therefore refer to this empirically supported phenomenon as harmful continuation. Beyond this intervention, we further characterize the removed post-conclusion continuation through uncertainty and hidden-state progress. We observe persistent local uncertainty together with weakened terminal-directional progress, forming an uncertainty--geometry mismatch. Finally, we instantiate Harmful Continuation Cut (HCC), a lightweight boundary proxy that approximates the editor-identified post-conclusion continuation boundary.","upvotes":6,"discussionId":"6a1fa59ee292c1c78ecb1372","ai_summary":"Answer-correct long chain-of-thought traces can lead to different fine-tuning outcomes, with post-conclusion continuations identified as harmful to training, characterized by uncertainty-geometry mismatches and addressed through a lightweight boundary proxy method.","ai_keywords":["chain-of-thought","large language models","supervised fine-tuning","answer-correct traces","post-conclusion continuation","delete-only editor","uncertainty","hidden-state progress","uncertainty-geometry mismatch","Harmful Continuation Cut"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct"},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"63369da91ba5d5ece24118a4","avatarUrl":"/avatars/67889e1ecadb04100a77bc8b5284c6fd.svg","isPro":false,"fullname":"wuyuhao","user":"mozhu","type":"user"},{"_id":"66a0d8cbc77099d013e9da92","avatarUrl":"/avatars/59f9d9529a246510c9f8e13c0dd6605c.svg","isPro":false,"fullname":"mantou","user":"mantou-cloud","type":"user"},{"_id":"684526dd103a93da4dc7d850","avatarUrl":"/avatars/5e756e1f5da6b68afb6c9ed165d3bc28.svg","isPro":false,"fullname":"Shenshen Li","user":"lss727","type":"user"},{"_id":"68d1117c32fe33a071b2d55b","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/UAvJVZtMcAUeoiauLvvbk.png","isPro":false,"fullname":"Chen He","user":"Dhragatis","type":"user"},{"_id":"646def60df618b303b419323","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/646def60df618b303b419323/JLJGYen4-5M8ivsLsSk0w.jpeg","isPro":false,"fullname":"Lei Wang","user":"demolei","type":"user"},{"_id":"64808f625409aa3e3bbbb7a4","avatarUrl":"/avatars/6647d2faca41205592211c5b4d69ce67.svg","isPro":false,"fullname":"J","user":"CFM-X","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2605/2605.29288.md"}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2605.29288 in a model README.md to link it from this page.

Cite arxiv.org/abs/2605.29288 in a dataset README.md to link it from this page.

Cite arxiv.org/abs/2605.29288 in a Space README.md to link it from this page.

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