nx_sigfixture.nx

nx_sigfixture.nx -- A PROCESS THAT DIES BY THE SIGNAL YOU NAME: THE ONE FIXTURE FOR EVERY WAIT-STATUS RULER. WHY (2026-09-05, measured on the promote path). nx_behaveprobe reported rc=0 for a gate the crash guard had killed with SIGSEGV, twice, and that receipt cleared a crashing gate for promote. The estate's rule already existed (wait_status_rc: a signal death reads 128+signal, never 0; nx_waitrc_gate proves it in-process with a forked child) -- the served probe was simply older than the rule. What the estate had no way to do was ASK A LIVE, REGISTERED RULER how it reads a signal death, because no registered subject dies by a signal on purpose. This organ is that subject: nx_behaveprobe, nx_job_run, /api/gate_run and every other consumer of the tr_run_capture family can be pointed at it and their rc read back. usage: nx_sigfixture <signal 1..64> -- prints one line, then kill(getpid(), signal) on ITSELF nx_sigfixture exit <code> -- prints one line, then exits with that code (the control arm) SIGTERM (15) is the honest signal: no guard catches it, so the wait status carries a TRUE signal death and a correct ruler reads 143 where the old accessor read 0. SIGKILL (9) is the watchdog class (the bounded capture maps its own kill to TR_ERR_TIMEOUT). A guarded fault signal (SIGSEGV 11) is caught by the default-on crash guard and surfaces as exit 139 -- that is the guard's contract and this fixture does not pretend otherwise: pass 11 and you measure the guard, pass 15 and you measure the ruler. The line it prints BEFORE dying exists so a capture is never empty by cause: an empty capture plus rc=0 is the shape of a tool that never ran, and this fixture must never be mistaken for one. If the signal is blocked or ignored the process is still alive afterwards; it says so and exits SF_EXIT_SURVIVED, never 0, so a ruler cannot be acquitted by a signal that never landed. license_tier: ORIGINAL expect_exit: 143