The ranking of entities in complex competitive systems based on pairwise interactions is a
standard problem in network science. This work uses the men’s professional tennis circuit
as a case study, modelling pairwise match outcomes as a directed network, to evaluate
different ranking methods. This assessment is challenging, as players’ intrinsic ability is not
directly observable or quantifiable in reality. Historical ATP data were analysed to extract
calibration parameters governing player lifecycles, aging, and the distribution of talent.
Based on these elements, a dynamic simulation model was developed to generate synthetic
tennis circuits in which the latent skill score of each player is known. Comparing the various
ranking metrics against this ground truth reveals varied performances. Centrality-based
methods, such as PageRank, struggle to identify the top-tier players of the circuit. The
probabilistic maximum likelihood approach based on the Bradley-Terry model rivals the
official ATP points system, although the latter remains, overall, the most robust method for
reconstructing the underlying player hierarchy.