Get Started
Home
Topics
Search
Library
Research questionHow can code-repair models fix bugs without unnecessarily rewriting surrounding code?A repair can pass correctness tests yet alter unrelated code, making review harder and weakening confidence that the original implementation was preserved.
AI
Code Generation & Program Synthesis
Evaluation & Benchmarks
LLM Pretraining & Post-training
Reinforcement Learning
Research Paper
Latest papersRecent research connected to this question, newest first.When Models Edit Too Much: On the Fidelity of Minimal Code EditsThe evidence concerns frontier LLMs repairing code with controlled AST-level corruptions across 400 BigCodeBench problems, where reference solutions provide known minimal patches. It measures excess edits, cognitive complexity, and performance retention, and examines preservation instructions and post-training approaches; conclusions are grounded in this benchmark and corruption setup.research paper · Sep 3, 2026
Related questions
How can vulnerability-fixing history be reused to detect and repair similar undisclosed flaws in other code?How can coding agents repair scientific software when domain guidance may mislead them?How can repository-level coding-agent benchmarks detect review-constraint failures beyond passing functional tests?How can teams identify which design choices reliably improve repository-scale refactoring agents without breaking behavior?