Pathfinding is a central topic in AI for games, with many approaches having been suggested. But comparing different algorithms is tricky, because design choices stem from different practical considerations; e.g., some pathfinding systems are grid-based, others rely on a navigation mesh or visibility graph and so on. Current benchmarks mirror this trend, focusing on one set of assumptions while ignoring the rest. In this work we present a new unified benchmark using data from the game Iron Harvest. For 35 different levels in the game we generate several complementary map representations (grid, mesh and obstacle-set) and we provide a common set of challenging instances. We describe and analyse the new benchmark and then compare several leading pathfinding algorithms that begin from different assumption sets. Our goal is to allow researchers and practitioners to better understand the relative strengths and weakness of competing techniques.