Abstract

Procedural Content Generation (PCG) is the use of automatic techniques to create content for games, varying from the generation of fundamental units such as textures or sound to complex environments and entire games. It can be used to provide diverse challenges for players every session, and to generate large amounts of content that would be unfeasible to design manually. Puzzle games are a popular PCG domain due to their combinatorial complexity. However, the selection of an adequate representation for PCG in puzzles remains an unanswered question, which cannot ensure that the best possible results are attained. This research aims to bridge the gap by reviewing different representations used for procedural puzzle generation. A systematic literature review is conducted, identifying 931 articles between 2020 and 2025. After a thorough selection process, 111 studies were used to answer three research questions related to the usage of representations in procedural puzzle generation. Our results include a classification of representations in "logical" (related to the topology of the search space and move generation) and "physical" (data structures used for implementation), along with the factors that influence their selection. We also extend current taxonomy in puzzle game classification to give further context in the utilization of certain PCG techniques in certain domains. We conclude that a significant gap exists in studies addressing the problem of representation selection. Future studies may utilize existing frameworks and explore established problems with novel approaches to enhance the process of generating puzzle game instances.