nx_team_out_lock_gate.nx

nx_team_out_lock_gate.nx -- REFEREE for the team_out.tsv concurrent-writer lock (WMS torn-write follow-on). Proves an rt_lock-serialized WRITE SESSION keeps a MULTI-LINE record block contiguous under concurrency, AND that the detector sees failure without the lock (mandatory neg-control). WHY a lock, not fa_appendz: express_lane writes a MULTI-LINE block (ACK lines + the SITREP block); per-line atomicity (fa_appendz) keeps each LINE intact but does NOT stop a second writer's lines from interleaving BETWEEN this writer's lines -> the block is shredded. The fix (this gate proves) is to hold rt_lock across the whole multi-line session so no other writer can slip a line in. GOOD lane: NWORKERS fork; each acquires rt_lock(GOODLOCK), writes NRECS records (each = L lines, each line emitted as a SEQUENCE of sys_write so it CAN tear), releases. Expect: ZERO torn lines AND ZERO block-contiguity breaks AND all N*NRECS*L lines present. BAD lane (NEG-CONTROL): identical, but NO lock -> concurrent workers interleave -> torn lines AND/OR block breaks MUST appear. If bad has none, the detector is worthless -> RED. A line is CLEAN iff it begins "BLK w=" AND the 4 bytes before '\n' are " END" AND exactly one "BLK w=" head AND its w field == its z field (the two wid copies bracket the line; an interleave splits them). A BLOCK BREAK = a clean line with s>0 whose immediately-preceding line is not the same (w,r) at s-1 (the block's L lines were not contiguous). Per-run unique /tmp paths. Self-log via the assembled single sys_write (fd is uncontended). GREEN(0) iff all hold. Sovereign: nx_registry_lock (rt_lock/rt_unlock; nx_syscalls transitive). license_tier: ORIGINAL