The article introduces a method for organizing complex data in a way that maintains the original order. By using hierarchical clustering techniques, the method creates a structure where if one item is ranked higher than another in the original data, their clusters will reflect this relationship. This approach combines traditional clustering methods with ultrametric fitting to find the best way to represent the data. The key idea is to create partial dendrograms that capture the connections between different components of the data. Ultimately, the goal is to produce a clustering that closely matches the original data's dissimilarity measure.