nx_flap_probe.nx

nx_flap_probe.nx -- TEAM-OWNED repeat-run stability probe (Engineer's verify verb). Runs a target ELF N times via fork/execve/wait4 and judges from the RAW wait status -- never a shell $? (the fake-green channel, see feedback-wsl-exit-code-judging-2026-06-09: $? echoes pre-expand through PS AND Git-Bash into wsl, so every inline RC=$? is the OUTER shell's last status, not the target's). Born 2026-06-10: a P3 KAT "passed" via $?-echo and "failed" in the game gate -- the gate (wait4) was right, the $? channel lied, and the hunt for "same-binary nondeterminism" was chasing a phantom. This organ makes the honest repeat-run judgment a one-call team capability, in pure NishiLang from the syscall layer up (no sh, no py). Usage: nx_flap_probe.elf <target.elf> [n_runs] (default n_runs = 20) Output: one FLAPPROBE line with a status histogram, e.g. FLAPPROBE target=/tmp/ft.elf runs=20 distinct=1 status=4 x20 verdict=STABLE FLAPPROBE ... distinct=2 status=0 x17 status=4 x3 verdict=FLAP Exit: 0 = STABLE (all runs same status; the status itself is REPORTED, not judged -- a stable failure is the caller's evidence, not ours); 2 = FLAP (nondeterminism proven); 3 = usage/exec error. license_tier: ORIGINAL