Human motion understanding and activity recognition have become popular tasks in recent years. The ability to recognize human motions and activities has been adopted in many application domains including sports. However, most of the existing prototype applications recognize and quantify the difference of motions holistically. The (dis)similarity of such action is typically expressed singularly in one number. In this work, the objective is to pursue a deeper structural understanding of the nuances in activities of the same type. Specifically, speed climbing performances will be analyzed by applying numerical and visualization-based methods that exploit movement patterns in estimated 2D skeleton data. The aim of the analysis will focus on the identification key aspects that most attribute to either a high degree of similarity or disparity of two compared performances. The outcomes of the analysis should be supported by human-friendly visualization, e.g., augmenting the observations directly into the split-screen or overlayed videos.