You run the Leiden community detection algorithm on your extracted graph, but it generates either one massive community (under-partitioning) or thousands of tiny, disconnected pairs (over-partitioning).
How does this community partitioning resolution affect your GraphRAG query performance and LLM token costs?