nx_sts_lock_gate.nx

nx_sts_lock_gate.nx -- proves sts_append_row's lock actually prevents the seq1559 lost-row race. THE COMPARISON IS CONTROLLED: both arms run the IDENTICAL read-modify-write with the IDENTICAL artificial window (a sleep between load and commit). The ONLY difference is whether <prefix>plock is held across it. If the locked arm keeps every row and the unlocked arm loses some, the lock is doing the work -- nothing else varies, so nothing else can be credited. NON-VACUITY IS THE POINT: a concurrency test that never triggers the race proves nothing at all. The unlocked arm MUST lose rows. If it does not, this gate reports VACUOUS rather than GREEN -- a guard that cannot be shown to fire has not been tested. license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0