GitHub has been tackling technical debt accrued over its rapid growth by refining and rewriting parts of its codebase to improve efficiency and maintain high engineering standards. A notable example is the recent replacement of a critical code path responsible for merges in pull requests. Historically, GitHub used a custom merge strategy due to limitations in Git's default merge tools, but these presented performance and cleanup challenges. To address this, GitHub leveraged libgit2, a re-implementation of Git's low-level APIs, to develop a more efficient in-memory merge engine that avoids disk usage. Transitioning to this new system involved extensive testing using Scientist, a library designed to compare old and new code paths in production without affecting users. This approach allowed GitHub to identify and resolve mismatches and performance issues efficiently, leading to a seamless switch to the new implementation. The effort not only improved performance but also led to the discovery and correction of bugs in both the original Git implementation and the new libgit2 setup. This systematic and experimental approach to refactoring ensures GitHub can manage technical debt effectively while maintaining robust and reliable service.