nx_schemadrift_gate.nx

nx_schemadrift_gate.nx -- proves the safety-flag ruler DISCRIMINATES (2026-08-25). The defect under test is a deriver that fails PERMISSIVE: it publishes readOnly=1 for an organ that reaches a write syscall through an imported helper. A gate for that ruler must therefore prove BOTH directions on EVERY axis it claims, because a scanner that answers "writer" to everything passes every permissive test while destroying all the information, and a scanner that answers "read-only" to everything passes every conservative test while destroying all the safety. THE LOAD-BEARING TOOTH IS T6. It runs the ruler at two depths over ONE REAL SUBJECT that ships in this estate -- nx_divprobe, which writes nothing itself and imports nx_gate_verdict.nx, whose gv_journal appends to knowledge/status/harness.jrnl. Depth 0 must say read-only and the transitive closure must say writer. A single-file scanner CANNOT pass it, which is the whole point: an anti-vacuity tooth the trivially wrong implementation fails by construction. T7 is the negative control that stops the opposite cheat. nx_syscalls.nx is imported by essentially every organ in the estate and it DEFINES every write syscall there is; if the ruler counted those definitions the entire estate would read as writers and the derivation would buy nothing. Its two occurrences of the mkdir marker are one COMMENT and one DEFINITION, so the shim alone must measure clean. Get this wrong and T6 still passes -- which is exactly why both are here. license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0