Tree Navigation Without LLM Summaries: A Matched-Cost Study of Hierarchical Retrieval for Long-Document QA

NavTree, a leaves-only hierarchical retriever, builds a deterministic balanced segment tree over document chunks with zero language-model calls at indexing and uses a hybrid lexical-and-dense frontier walk for navigation, emitting only leaf chunks to the reader. On a matched-cost evaluation, NavTree is the strongest matched-cost hierarchical retriever and ties the strongest flat baseline, demonstrating that summary trees' main benefit in long-document QA comes from navigation rather than generated summary content.