nx_sts_cas_gate.nx

nx_sts_cas_gate.nx -- does an UNLOCKED sts_seed writer LIE ABOUT SUCCEEDING? nx_sts_lock_gate already proves the lost-row race exists (1 of 6 rows survive unlocked). It counts SURVIVING ROWS, and that is the one thing it cannot use to tell the two outcomes apart: SILENT CLOBBER : sts_seed returns a row count, the caller believes it wrote, and the row is gone. LOUD REFUSAL : sts_seed returns negative, the caller KNOWS it did not write, and can retry. Both leave fewer rows in the plane. Only one of them is a data-loss defect; the other is ordinary backpressure. So this gate measures the thing that actually matters: SILENT LOSS = (writers that were TOLD they succeeded) - (rows actually present) Pre-guard that is 5 of 6: six writers each get a success return, one row exists. Every one of those five callers has already moved on, logged success, and returned a row id nobody will ever read. A CAS/generation guard cannot make more rows survive -- the writes genuinely conflict -- but it can drive SILENT LOSS to ZERO, which converts an invisible corruption into a retryable error. NON-VACUITY: if every writer succeeds AND every row survives, nothing raced and the gate says VACUOUS rather than GREEN. A concurrency gate that never triggers the race proves nothing. license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0