Recommender systems are versatile. Algorithms such as collaborative filtering, DeepFM and Deep Matrix Factorization can be deployed in different commercial environments without a lot of fundamental modifications. Although there has been hot debate among academic scientists and industrial product owners, environment-sensitive algorithms remain a practice only in real world applications. In this paper, we discuss the topic whether collaborative filtering algorithm is a justified valid model by topological analysis of the problem structures. We claim that the validity of the algorithm is related to the shape of the input data domain.