A companion article (Article 7quater) in the Enterprise Document Intelligence series addresses hierarchical retrieval for documents too long to read, with a table of contents too long to dump. The problem is demonstrated using NIST SP 800-53 Rev. 5, a 492-page document defining security controls. A user asking "What does the account management control require?" expects the answer on pages 46 to 50 (control AC-2). Naive RAG, which embeds every page and takes the top-k most similar, fails because the words "account, management, control, and access" appear on hundreds of pages, returning AC-2 mixed with other controls and the glossary. The table of contents itself has 358 entries. The solution mimics an expert who scans the chapter list (eleven titles), picks "The Controls," opens the twenty families, picks "Access Control," and lands on AC-2. The article builds a loop that makes one small decision at a time, amplifying the expert rather than dumping the whole document or table of contents on the model. Runnable companion notebooks are on GitHub at doc-intel/notebooks-vol1.