nx_deep_research_test.nx

nx_deep_research_test.nx -- the team does what /deep-research does, minus the LLM fetch+extract, to SAVE TOKENS. For a typical run (5 angles, 25 claims, 3 sources/angle), the full-LLM pipeline is ~108 agent calls; the team owns scope+verify+synth+task (incl. the 75-call verify, done deterministically), leaving only fetch+extract (~30) for Claude -> ~72% token saving, AND the verify is now reproducible. Exit 0 if the accounting holds. license_tier: ORIGINAL