nx_ws_crash_gate.nx

nx_ws_crash_gate.nx -- WMS-R10: the END-TO-END CRASH INTEGRATION TEST (the capstone). This is the scenario that STARTED the whole arc -- "claude froze and crashed, find out where we were" -- proven a NON-EVENT. It composes every durability rung at once: R0/R0b framed-append floor : concurrent workers record transitions to the REFLOG with zero tear R2 transition ledger : the append-only reflog = the durable source of truth R8 ws_put_locked : concurrent workers also update the LIVE registry, no lost updates R9 ledger_rebuild : after the crash WIPES the registry, the reflog rebuilds it FLOW: NWORKERS fork; worker w writes NTRANS transitions for workstream w to BOTH the reflog (ledger_append) AND the live registry (ws_put_locked), ending at a deterministic terminal state. Then we simulate a CRASH that loses the derived registry (a fresh, empty registry prefix) while the append-only reflog survives -- the realistic failure mode. Recovery = ledger_rebuild(reflog -> new registry). The end-to-end claim: EVERY workstream's state is recovered, NOTHING is lost. ASSERTIONS (no false green): T1 floor-held : reflog has exactly NWORKERS*NTRANS lines AND flagged==0 (R0 held under concurrency -> the history we recover from is complete and untorn). T2 live-consistent : the concurrently-built LIVE registry has all NWORKERS workstreams at their correct terminal state (R8 lost no update across the shared manifest RMW). T3 recovery-complete : after the crash (wiped registry), ledger_rebuild restores all NWORKERS workstreams with correct terminal states (recovered==N AND correct==N). T4 load-bearing(NEG) : a wiped registry that is NOT rebuilt has ZERO of the N workstreams -- so the recovery is what saves them; if this showed N, the test would be vacuous. GREEN(0) iff T1..T4. Self-log via fa_appendz (dogfoods R0b). Evidence -> ws_crash_gate.log. Sovereign: nx_ws_repair + nx_ws_cas + nx_ws_ledger + nx_workstream_store + nx_framed_append + nx_syscalls. license_tier: ORIGINAL