nx_research_digest.nx

nx_research_digest.nx -- end-to-end EXTRACTIVE research digest (sovereign, no LLM). Given a query + a set of sources (id, class, text) + candidate claim terms, it returns: - sources RANKED by BM25 relevance to the query (out_src_order) - each claim's CONFIDENCE by independent corroboration (out_claim_conf) - the corroboration ledger (caller reads citations via nx_corrob_cite) A source "asserts" a claim iff its text contains the claim term (nx_bm25_tf). Corroboration uses the independence rule (nx_research_corroborate): echoes of one source-class don't add confidence. This is the rigorous, cited digest the Nishi researcher can return TODAY -- the generative prose synthesis on top is the LLM-gated step (ai/accel lock). Composes nx_bm25 + nx_corroborate (DRY #15). The live FETCH that supplies the sources = nx_crawl_https (the delegated step), wired next. license_tier: ORIGINAL