一句话：提出“多补丁指纹”，把来源信息分散嵌入修复补丁组合中，勾勒可问责自动程序修复的研究议程。
基于大语言模型的自动程序修复工具已能大规模生成补丁，却造成问责真空：AI 补丁一旦出错，责任难以归属，而现有修复系统缺乏验证补丁来源的机制。本文论证可追溯修复的必要性，指出现有数字水印方案因补丁必须正确、最小且自然等约束而难以直接适用，进而提出“多补丁指纹”：不把来源信息压入单个补丁，而是通过行为签名与机会式嵌入将其分散到整个修复补丁组合中。文章剖析修复约束与水印之间的张力，并给出面向可问责修复生态的研究议程。作为立场与议程论文，本文未报告实证评估。
Automated Program Repair (APR) has evolved from academic curiosity to industrial necessity. Large Language Model-based repair tools now generate patches at unprecedented scale, yet this deployment creates an accountability vacuum: when AI-generated patches fail, who bears responsibility? Current APR systems lack mechanisms to verify patch provenance—a critical gap as regulations demand transparency and organizations struggle with liability attribution. We propose that watermarking may offer a path toward accountable repair, but APR imposes unique constraints: patches must be correct, minimal, and natural. These constraints render existing watermarking insufficient. We propose multi-patch fingerprinting, distributing provenance across repair portfolios rather than individual patches through behavioral signatures and opportunistic embedding. This paper articulates why traceable APR matters, examines tensions between repair constraints and watermarking, and outlines a research agenda toward accountable repair ecosystems.