nx_commons_defect_gate.nx

nx_commons_defect_gate.nx -- proves, against the REAL shipped organs, that the estate's pairwise reciprocity audit SILENTLY STOPS DETECTING EXTRACTION once the flows get large. WHY THIS CALLS THE REAL FUNCTIONS AND NEVER A COPY: a hand-copied differential probe is evidence only if proven character-identical, and this estate has already paid for that lesson -- a copied probe silently FIXED the bug it was meant to expose and exonerated an innocent compiler for four rounds. So this gate imports nx_mycorrhiza.nx and calls nx_my_audit_kind on a real NxMycorrhiza built by the real constructor. If the audit is fixed, this gate goes green without being touched. THE DEFECT (read from source, then made to fire here): nx_my_audit_kind computes (native_gives_q * NX_MAGIC_1024) / foreign_gives_q. native_gives_q * 1024 overflows i64 above ~9.0e15 units. On overflow the product goes NEGATIVE, every `> tolerance` test is false, both net-giver/net-taker tests are false, and the function falls through to OK_BALANCED. IT FAILS TOWARD "NO EXTRACTION DETECTED", AND IT DOES SO ON THE LARGEST FLOWS FIRST. 9.0e15 is 9 PB of DISK_BYTES, or 285 CPU-years of CPU_MICROSECONDS -- reachable by a fleet, and reachable soonest by exactly the actor moving the most volume. T1/T3 are the NEGATIVE CONTROLS: the same extraction at ordinary scale MUST still be caught. If T1/T3 ever fail the fix broke the detector rather than extending it. license_tier: ORIGINAL expect_exit: 0