This article is a description of the continuation of previous research by the authors related to the restoration of distance matrices. The main difficulty that arises with such a recovery is that it is impossible to use conventional techniques such as variants of gradient descent algorithms, since too many variables would arise. In addition, in this article we also do not use algorithms for variants of the branch and boundary method, which in our previous publications were sometimes used for the problem considered in the article; the main argument for not using it is that using it would provide little information about the subtasks generated by this algorithm, in particular, it is difficult to adequately assess the real the values of the resulting boundaries. Therefore, we use the so-called step-by-step filling of the distance matrix. In the approach presented in the paper we consider, that the algorithm of the initial distance formation works in the best way. Therefore, we conditionally believe that the corresponding value of badness cannot be improved. This assumption really makes it possible to improve the value of badness. Thus, in this paper we have obtained practical results of reconstructing distance matrices, which significantly improve the results given in our previous papers.