nx_deep_research.nx

nx_deep_research.nx -- build the RESEARCHER up to do what /deep-research does, so the team owns the mechanizable bulk and Claude only intercedes at the bounded FETCH+EXTRACT (operator: save tokens). A /deep-research run is ~108 agent calls across 6 stages; the heavy one is VERIFY (3-vote over ~25 claims = 75 calls) -- and that is DETERMINISTIC, so the team does it sovereignly (and reproducibly, which is BETTER than a varying LLM vote). Of the 6 stages, 4 are TEAM-owned (scope, verify, synthesize, task -- nx_researcher already does verify+synth+task) and only 2 are LLM-gaps (fetch external sources, extract prose->claims). The orchestrator runs the pipeline and accounts the token saving. RACI: Researcher orchestrates + owns the mechanizable stages; Claude does fetch+extract; Librarian persists. license_tier: ORIGINAL Refs: nx_researcher (verify/synth/task); the LLM-gap registry.