The paper examines the performance of modularity maximization, a popular technique for identifying modules in networks, in practical contexts. It revisits and clarifies the resolution limit phenomenon, demonstrating that the modularity function $Q$ exhibits extreme degeneracies, with an exponential number of distinct high-scoring solutions and often lacks a clear global maximum. The authors derive the limiting behavior of the maximum modularity $Q_{\max}$ for infinitely modular networks, showing its dependence on network size and the number of modules. Using real-world metabolic networks, they show that degenerate solutions can disagree on many partition properties, such as the composition of large modules and module size distribution. These findings suggest that the output of modularity maximization should be interpreted cautiously in scientific contexts and highlight the need for heuristics to find high-scoring partitions. The paper also discusses methods to mitigate these issues, such as combining information from multiple degenerate solutions or using generative models.The paper examines the performance of modularity maximization, a popular technique for identifying modules in networks, in practical contexts. It revisits and clarifies the resolution limit phenomenon, demonstrating that the modularity function $Q$ exhibits extreme degeneracies, with an exponential number of distinct high-scoring solutions and often lacks a clear global maximum. The authors derive the limiting behavior of the maximum modularity $Q_{\max}$ for infinitely modular networks, showing its dependence on network size and the number of modules. Using real-world metabolic networks, they show that degenerate solutions can disagree on many partition properties, such as the composition of large modules and module size distribution. These findings suggest that the output of modularity maximization should be interpreted cautiously in scientific contexts and highlight the need for heuristics to find high-scoring partitions. The paper also discusses methods to mitigate these issues, such as combining information from multiple degenerate solutions or using generative models.