nx_arbiter_race_gate.nx

nx_arbiter_race_gate.nx -- MEASURED before/after proof that nx_arbiter turns F-class coordination into S-class on a REAL shared-sink race. RG_N child PROCESSES (sys_fork -- our own kernel clone, no shell) each do RG_K iterations of: read counter file -> increment -> write back. That read-modify-write is EXACTLY the pattern that clobbered MEMORY.md + genesis_lineage.tsv mid-write today. A 0.5ms window between read and write forces the scheduler to interleave the children. UNLOCKED : writers race -> updates are lost -> final count << RG_N*RG_K. ARBITRATED: fl_acquire serializes the critical section -> final count == RG_N*RG_K, exactly, every run. Self-contained + sovereign: our fork/wait4 + our flock-based arbiter + our counter file. No 3rd party, no shell-driven race. GREEN iff (a) the unlocked race PROVABLY loses updates AND (b) the arbiter eliminates every one of them. This is the honest measured exceed -- a number, not a claim. license_tier: ORIGINAL