nx_dr_chain.nx

nx_dr_chain.nx -- the CAPSTONE: one call runs the whole deep-research loop (DR-CHAIN). Composes the live DR organs end-to-end instead of leaving them as eight separate calls: JUDGE nx_dr_run tokenize each source doc, token-containment vs a SHARED groundtruth-insight set (the DRBench task shape) -> coverage CONDUCT nx_dr_conductor verify-yield -> Elo Tournament-of-Ideas -> opportunity-cost budget dispatch -> which sources earn deep compute SCORE nx_dr_drbench insight-recall over what was actually selected ★NOVEL MEASUREMENT this makes possible: recall_all (if you could read every source) vs recall_selected (what the budget-constrained agent actually got) -- the MEASURED COST of the opportunity-cost constraint, which no single-agent tool reports. Packed self-describing inputs keep every func <=6 params (NAS nx_cc >6-arg skew, seq239). Deterministic/integer. No hardware writes (Rule 26). module: nishi-core.research.dr_chain depends: nx_dr_run.nx, nx_dr_verify.nx, nx_syscalls.nx genealogy_id: coscientist_2026_pipeline + drbench_2026_task_shape