nx_library_search_test.nx

nx_library_search_test.nx -- the Researcher handles a LIBRARY OF INFORMATION, not one source, and searches it with S-class ranked retrieval. End to end, sovereign: 1. FETCH the index, RECOGNIZE it is a library (>=3 [doc] entries) 2. HARVEST every listed doc -- follow each entry, fetch it, MIRROR it into the Nishi Library 3. build the corpus FROM THE LOCAL MIRRORS (bot-block-proof) 4. answer 3 different SEARCH PROMPTS by TF-IDF RANKING the whole corpus -- each returns the single best-matching doc, where naive substring search would be ambiguous (e.g. "abandon" is in 2 docs) Exit 0 on 8/8. license_tier: ORIGINAL NOTE 2026-06-20: this is a LIVE-INTEGRATION test -- it requires a fixture HTTP server on 127.0.0.1:8099 serving a library index + the 3 docs. With no server up it reads 0/8 ("ambiguous across 0 docs") because the harvest fetches nothing -- that is an unmet fixture dependency, NOT a search-logic regression. The REPRODUCIBLE-BY-OUR-SYSTEM proof of the ranked-retrieval claim (no socket, real on-disk corpus, both TF-IDF + production BM25) is nx_library_search_gate (8/8 GREEN). Run THIS one only with the :8099 fixture live.