In this talk, the speaker introduces a series of recent and ongoing research topics in the area of software repository mining and software fault detection. Mining software repositories has become an essential approach to empirically study software engineering practices, as it enables researchers and practitioners to analyze large-scale, real-world development data. Building upon this perspective, the talk revisits several past studies, including the impact of feature reduction techniques on defect prediction models, an empirical study of issue-link algorithms that connect bugs and commits, the benefits and pitfalls of token-level SZZ in identifying bug-introducing changes, and an empirical study of token-based micro commits. These works highlight both methodological challenges and practical implications in handling fine-grained software artifacts. Furthermore, the speaker presents more recent progress, with a focus on two approaches for improving faulty interaction localization using logistic regression analysis. These approaches aim to enhance the accuracy and efficiency of identifying fault-inducing interactions in software systems, thus contributing to more reliable defect detection and debugging support. The significance of these studies lies in their potential to improve the quality and maintainability of software systems, which are increasingly critical to society. As modern software projects grow in complexity and scale, traditional defect detection methods often fail to capture subtle yet important sources of faults. By advancing techniques such as fine-grained commit analysis, improved issue-linking, and statistically grounded localization models, this line of research contributes to reducing maintenance costs, preventing system failures, and ultimately enabling developers to deliver more robust software. Through this talk, the audience will gain insights into not only the technical contributions but also the broader impact of empirical research on the future of software quality assurance.