nx_sendcheck_gate.nx

nx_sendcheck_gate.nx -- LN6 (lang.plan): DATA-RACE TYPING AT THE THREAD-POOL BOUNDARY. End-to-end over the PROMOTED compiler: four fixtures are assembled at runtime in /tmp/<gate>/ (never a file a source census could mistake for an organ), the compiler is forked on each with stderr captured, and the teeth read the captured bytes. Each fixture declares its own nx_pool_submit stub (the check is keyed on the callee NAME, so the real pool and its threads are not needed): fx_unmarked `struct Job { .. }` handed as `j as i64` under --sendcheck -> MUST be refused, naming Job, with where (caret), why and fix (the teaching voice) fx_marked `struct Job send { .. }` the same call under --sendcheck -> MUST compile (the marker works) fx_int a plain integer ctx under --sendcheck -> MUST compile (declared abstention) fx_default fx_unmarked WITHOUT the flag -> MUST compile (default unchanged) The refusal is the LN6 contract; the three controls keep the gate from passing on a compiler that refuses everything, on one that ignores the marker, or on one that changed the default. Byte-identity of default builds is nx_cc_equiv_gate's job (10 rows + self-host), deliberately not duplicated here. nx_sendcheck_gate -- verdict in the exit code (gv_verdict) license_tier: ORIGINAL layer: lang module: nishi-core.lang.sendcheck_gate