In Surgical Data Science (SDS), there is an increasing demand for large, realistic annotated datasets to facilitate the development of machine learning techniques. However, in laparoscopic surgery, most publicly available datasets focus on low-granularity procedural annotations (such as phases or steps) and image segmentation of instruments or specific organs, often using animal models that lack clinical realism. Furthermore, annotation variability is seldom evaluated. In this work, we compiled 30 sleeve gastrectomy procedures and performed three levels of annotations for a specific step of this procedure (the fundus dissection): a procedural annotation of fine-grained activities, a semantic segmentation of the laparoscopic images, and the assessment of a surgical skill, specifically the quality of exposition of the surgical scene. We also conducted a comprehensive annotation variability analysis, highlighting the complexity of these tasks and providing a baseline for evaluating machine learning models. The dataset is publicly available and serves as a valuable resource for advancing SDS research.