nx_gatebuilt_gate.nx

nx_gatebuilt_gate.nx -- THE UNBUILT-GATE CENSUS. For every authored *_gate.nx, does a promoted *_gate.elf actually exist? A gate that lives only in source makes its subject PERMANENTLY UNVERIFIABLE, and nothing in the ecosystem reports it. WHY (found by accident 2026-07-30, id 1785472297): I tried to VERIFY the FD_CLOEXEC port-hostage fix before eating three sev-9 rows that depend on it. `POST /api/gate_run target=nx_cloexec_gate` returned **exit_code=127 verdict=NOT-FOUND bytes=0** -- a fully-written gate (T1 control proving the disease state is reachable, T5 neg-control, T6 non-vacuity) that had NEVER BEEN COMPILED. Building it took one call and it came back 6/6 GREEN. Until then, three sev-9 incidents rested on a fix nobody could prove. A two-glob spot check then said the gap is systemic: ~2877 *_gate.nx sources vs ~187 *_gate.elf binaries. WHY NOTHING ELSE CATCHES IT -- an unbuilt gate is invisible in EVERY direction: nx_gate_rollup reads LOGS -> no binary means no log means the gate is ABSENT from the roster, not RED. nx_wirecensus S3 counts promoted-but-unregistered -- never authored-but-unbuilt. every debt row citing 'gate-proven' reads as evidence regardless. It is the exact INVERSE of PROVEN_UNWIRED (4952 capabilities BUILT+GATED+never called): there the gate exists and the capability is unused; here the capability is used and the gate does not exist. BOTH SCORE AS DONE. LAW: AN UNBUILT GATE IS INDISTINGUISHABLE FROM A PASSING ONE IN EVERY REPORT THAT MENTIONS IT. SECOND CLASS, ADDED 2026-08-07 -- STALE-BUILT: the binary EXISTS but PREDATES its own source. Strictly WORSE than unbuilt, because unbuilt fails LOUD: /api/gate_run answers exit 127 NOT-FOUND for a missing binary, but exit 0 verdict=GREEN for a stale one -- and a GREEN is believed. Found the expensive way: nx_fsops_gate served pass=23/23 GREEN off a 2026-07-29 binary while its source declared 28 checks. The contradiction sat in the SAME response (printed total != declared total) and nothing read it; I reported "no regression" on that GREEN. MEASURED on this census first run: 3168 sources, 715 built, of which only 176 are CURRENT and 539 are STALE-BUILT -- so 5.6pc of authored gates can speak about their own source. LAW: A STALE GATE IS WORSE THAN AN UNBUILT ONE. UNBUILT FAILS LOUD; STALE PASSES QUIETLY. COROLLARY, and the reason this organ must be rebuilt on every edit: EDITING A GATE SOURCE WITHOUT REBUILDING IT CREATES THE VERY CONDITION THIS GATE REPORTS. ENVELOPE, declared in the output and never silent: FLAT scan of the two named dirs (no recursion, so the _attic/_retired/_stage_local shadow trees are deliberately EXCLUDED and this counts LIVE gates only); existence is probed with openat, not a read, so it is cheap and cannot be fooled by an empty file being unreadable; the shown list is capped and the cap is printed. license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0