nx_osbench_kern.nx

nx_osbench_kern.nx -- rung F853b (lane osbench): the KERNEL-AXIS harness. WHY THIS EXISTS: the operator mission is RAW KERNEL PERFORMANCE, and 750 of 1000 permille of that mission (syscall_trap, ipc, thread_spawn, context_switch) was UNMEASURED -- we did not even know what LINUX costs on this hardware, so we had no target to aim at. This organ measures the LINUX side of the kernel axes so the number NishiOS must beat is on record. ★ HONEST SCOPE, STATED UP FRONT: these are REFERENCE rows, not comparisons. NishiOS does not run as an OS on this substrate -- it is hosted, so it has no syscall-trap or scheduler of its own here. There is therefore NO NishiOS counterpart to pair against, and the judge scores NOTHING from this file. Emitting a one-sided number as if it were a head-to-head would be exactly the no-wave violation this lane exists to prevent. The NishiOS halves stay UNMEASURED, with the rung that unblocks them named (F856 bare-metal node). ★ A STRUCTURAL NOTE THAT MUST NOT BE SCORED AS A WIN: NishiOS's native contract has no syscall trap on the store path at all (capability call, not a ring transition). That is a FLOOR/design difference, and the gap ledger's rule applies -- "we own it" is never printed as "we beat it". It is recorded as prose here, never as a ratio. AXES (Linux, this host): syscall_trap -- minimal round trip into the kernel and back (close on an invalid fd: a real trap, EBADF immediately, near-zero kernel work = trap floor) pipe_write_read -- a byte through a real pipe buffer, same process. NAMED HONESTLY: this is the pipe-buffer cost, NOT an inter-process IPC round trip and NOT a context switch. Calling it ipc_roundtrip would be a lie by label. license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0