One of our approaches, in this project, is to build verifiable SWE tasks based on public repository histories in a synthetic approach with human-in-the-loop; while expanding the dataset coverage to different types of tasks in terms of programming language, difficulty level, and etc. We are looking f