nx_framed_append_gate.nx

nx_framed_append_gate.nx -- the REFEREE for WMS-R0 (atomic framed-append). Proves the nx_framed_append primitive defeats concurrent-appender tearing, AND proves the test can actually DETECT tearing (the mandatory negative control). GOOD lane: N workers fork concurrently, each emits NRECS records via the NEW single-write primitive fa_append(GOODPATH, ...). Expect ZERO torn lines and ALL N*NRECS lines present (nothing lost, nothing torn). BAD lane (NEGATIVE CONTROL): the SAME concurrency, but each record is emitted the OLD way -- a SEQUENCE of separate sys_write() calls per record. O_APPEND only makes ONE write atomic, not the sequence, so workers interleave -> torn lines MUST appear. bad_torn == 0 => the detector is worthless => RED (no false green). TAMPER: an oversized record (rec_len+1 > cap) must be REJECTED with -2, never silently written/torn. A line is WELL-FORMED iff it starts "FA w=" AND the bytes just before its '\n' are "END" AND it contains exactly one " END" token. Any interleaving violates one of those -> counted as torn. Output: rows -> stdout + the evidence log knowledge/status/framed_append_gate.log, then a final verdict line. Exit 0 on GREEN, 1 on RED. Sovereign: only nx_syscalls + nx_framed_append. license_tier: ORIGINAL