nx_netsync_gate.nx

nx_netsync_gate.nx -- CERTIFICATION of nx_netsync (gamebench gap-queue rank 1: networking-multiplayer). The existing nx_lockstep/nx_rollback gates prove netcode mechanisms on a TOY stepper; this gate proves the same laws hold when the simulation is the REAL certified world (nx_worldsim) and the snapshots are REAL save files (nx_gamesave) -- the composition the capability actually demands. T1 LOCKSTEP DETERMINISM -- two peers, same commands, different within-tick arrival order, 200 ticks -> checksums equal EVERY tick and shared state word-exact T2 STALL-NOT-GUESS -- missing one peer's input for a tick freezes the session (no advance, no state change) until it arrives; advancing on partial input is THE lockstep desync bug T3 ARRIVAL-CHAOS IMMUNE -- the full 400-command schedule delivered in a shuffled order with duplicates -> final state EXACTLY equals the in-order run; a conflicting resend is REFUSED T4 DESYNC LOCALIZED -- a 1-unit wealth mint injected into one peer at tick 120 is first visible in the checksum stream AT tick 120 (never earlier, never missed) T5 ROLLBACK ON THE REAL SIM -- speculative head with predicted inputs diverges when the prediction was wrong; discard + resim from the confirmed state converges bit-exact to ground truth; a CORRECT prediction triggers no divergence (no false rollback) T6 LATE-JOIN VIA nx_gamesave -- a peer that loads a mid-session snapshot file into a ZEROED arena tracks the original checksum-for-checksum to the end (GX-9 transparency) T7 BOUNDED + LOUD -- out-of-range tick/peer and conflicting input each refuse with a DISTINCT code and change nothing T8 ANTI-VACUITY -- the session actually traded and grew (a frozen or all-no-op world would pass T1-T3 while proving nothing) T9 ARTIFACT -- the final session round-trips through a real save file and RESUMES identically; knowledge/nx_netsync_session.sav is the board evidence license_tier: ORIGINAL expect_exit: 0