nx_stderronly_gate.nx

nx_stderronly_gate.nx -- END-TO-END gate for nx_stderronly, as an ORGAN. No shell. Inherits nx_gate_verdict (gv_ctr/gv_head/gv_check/gv_bite/gv_verdict) so the EXIT CODE CARRIES THE VERDICT and the declared count equals the executed count by construction. Runtime plumbing (fixture writing, fork/exec, output capture) comes from nx_gatekit_lib. EVERY FIXTURE IS ASSEMBLED AT RUNTIME UNDER /tmp/nx_stderronly_gate/ AND THE QUOTE CHARACTERS ARE COMPOSED FROM A BYTE, NEVER WRITTEN AS LITERALS. The subject is a SOURCE SCANNER, so a fixture spelled out in this file would be found by the subject when it sweeps the corpus -- a detector that scans source will find its own test fixture, and writing the pattern in a COMMENT re-arms the same trap because prose is source bytes too. A gate must also not share its fixture with a production beat, which is why the ratchet teeth pass --baseline into /tmp rather than touching nishi-ops. THE ANTI-VACUITY TOOTH IS FIRST AND IT IS A gv_bite: a trivial always-GREEN implementation fails the fires-on-bad half and a trivial always-RED implementation fails the silent-on-good half, so neither can pass this gate. Every refusal-side tooth is named neg-control-... so the census can SEE it. POSITIVE CONTROLS ON REAL ESTATE CODE, because a detector that flags a known-good guard is worse than none: nx_clockjob.nx (SEVEN real fd-2 refusal guards, every one of them carrying sys_exit) must be EXAMINED and NOT flagged, and nx_logtail_gate.nx must not be flagged. ⚠ THE SECOND CONTROL IS DELIBERATELY WEAK AND IS LABELLED AS SUCH: nx_logtail_gate contains no fd-2 write at all, so it could not be flagged by any implementation, which is why the clockjob control is the load-bearing one -- it has the emissions AND the refusal words AND must still come back clean. license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0