nx_boundedprobe.nx

nx_boundedprobe -- DOES THE ESTATE'S BOUNDED RUNNER ACTUALLY BOUND A SUBJECT THAT FORKS? gk_run_capture_ms exists because a hanging subject starved three estate sweeps (2026-08-07). Its watchdog kills the direct child with nx_kill(pid, 9). But the child's stdout/stderr are dup3'd onto a PIPE, and any grandchild it forks INHERITS those descriptors. Kill the child alone and the grandchild still holds the write end, so the parent's blocking read never sees EOF -- the watchdog fires, the child dies, and the sweep starves anyway. That is a hypothesis. This organ is the experiment, and it is self-contained so there is no /bin/sh on the path and no second organ to keep in sync: --forkling : fork a 60s sleeper, then exit IMMEDIATELY (the direct child is gone in ~0ms) (no args) : run MYSELF as that subject under a 2000ms bound, and report what came back Wall time is measured by the caller. ~2s => the runner bounds a forking subject. ~60s => it does not, and every gate in the fleet that runs a forking subject is exposed to the same starvation. license_tier: ORIGINAL No hw writes (Rule 26). Read-only apart from its own child process.