nx_res_census.nx

nx_res_census.nx -- THE RUNTIME RESOURCE WITNESS (per-daemon + system, trend-logged). 2026-07-05 OOM arc + operator: "efficiency in our resource use across ALL the nishi ecosystems ... live resources and storage and all the other axes" + "measured every poll, not discovered at OOM". WHY THIS ORGAN EXISTS (no-tool-proliferation: consolidation-check DONE, justified NEW): nx_explore for VmRSS / proc_status / resource_census / diskspace = 0 files -- NOTHING reads per-process memory anywhere in the ecosystem; the 26GiB health_eval leak + the 2.5TB clock commit were discovered AT OOM, and the leak diagnosis lived only in a memory note. This is the RUNTIME half of the two-witness law: the static half (nx_antipattern_catalog leaksweep) catches the DIRECT mmap-in-loop shape but proved BLIND to the helper-indirection tier (nx_seed_announce_all: 199MB grown, ZERO static sites) -- only a per-daemon RSS/VmSize GROWTH TREND catches that class. Single responsibility (Cardinal 9): MEASURE + LOG + show DELTA. It is a WITNESS, not a judge -- no thresholds baked in (Cardinal 11); the judge tier composes later (nx_metric_ledger BETTER/WORSE, mgmt_snapshot surface, clock-registry cadence = OWNER wiring on the NAS; hardware-up honesty: this run proves the mechanism on the dev substrate, the target wiring is hostctl/clock). SIGNATURE TABLE (banked by the OOM workstream): huge VmSize = commit-leak / fork-wedge class; huge anon-RSS = the OOM-killer class. Both columns logged so the trend separates them. MMAP DISCIPLINE: every buffer HOISTED once in main -- this organ must scan CLEAN under the leaksweep it complements (practice the law). license_tier: ORIGINAL lineage_id: nishi_res_census_v1 expect_exit: 0