Worked on the cvat-ai/cvat repository to enhance data integrity within labeling workflows by addressing a key bug related to annotation source alignment with label categories. Using TypeScript and front end development skills grounded in object-oriented programming, implemented a targeted fix that ensures the annotation source updates correctly whenever label categories change. This approach prevents inconsistencies in object properties during label updates, mirroring the established method for handling position changes. The work focused on improving the robustness and reliability of labeling data, reducing the risk of downstream errors in export and training pipelines, and laying a more stable foundation for future feature development.