you got it. not feasible to precompute the fine detail full graph, since it has far too many vertices, arguably feasible for the chunk-graph, since it is small.

you got it. not feasible to precompute the fine detail full graph, since it has far too many vertices, arguably feasible for the chunk-graph, since it is small. there'd still be search cost at runtime doing the fine-detail searches since the distance heuristic defined by the chunk-graph is an approximation and misleads the fine-detail search some of the time. need to profile & run experiments to see if this was anything like a win. in practice it might be better to increase the accuracy of the chunk-graph heuristic by making it consider more factors, which would make it harder to precompute and cache everything, but might further reduce the number of nodes the fine-detail search needs to expand edit: another variation of this would be to just start with an empty cache for chunk-graph distance matrix and fill it incrementally at runtime the time whenever you run a chunk-graph distance query as part of the heuristic. Then there isn't any up front time to precompute anything. Might be better if (source_chunk, dest_chunk) pathfinding queries at runtime aren't uniformly distributed but somewhat correlated, which is plausible, since some chunks are likely to be far more interesting locations to travel from/to than others.