nx_onsite_index.nx

nx_onsite_index.nx -- SITE-AGNOSTIC INGESTOR for the reusable onsite search client. Turns a site's search source (a 2-column "id<TAB>text" TSV -- the universal "search manifest source") into the durable pair the client reads: an inverted index (nx_search_inverted) + a MANIFEST (url<TAB>title<TAB>text, one line per doc, docid = line). REUSE not reinvent -- the gallery ALREADY produces id<TAB>text (knowledge/status/galx_search.tsv via nx_galx_search_index); any site emits the same shape and gets BM25 onsite search for free. "find files and articles": id = the file/article id, url = <url_prefix><id>, text = its searchable body. Usage: nx_onsite_index <src_id_text.tsv> <url_prefix> <idx_out> <manifest_out> [max_docs] e.g. nx_onsite_index knowledge/status/galx_search.tsv /img/ knowledge/index/gallery.idx knowledge/index/gallery.manifest COLLECT-THEN-BUILD with DEDUP: one pass collects the UNIQUE-by-id rows (galx_search.tsv had ~38% duplicate cids -- re-indexed images -- which would otherwise surface the same image many times); then a clean 2-pass index build + manifest write over the deduped set keeps idx docids and manifest lines aligned by construction. Pure-numeric tokens (image seeds/dims/steps) are scrubbed from the INDEXED copy (search noise); the manifest keeps the original text. license_tier: ORIGINAL