nx_mgmt_call_gate.nx

nx_mgmt_call_gate.nx -- TEETH FOR THE CONTROL-PLANE DRIVER FAILURE PATH, PROVEN WITHOUT CAUSING A FAILURE. WHY THIS EXISTS: nx_mgmt_call diagnostics were invisible to every MCP caller because they went to fd 2 while the tools-API capture surfaces fd 1. Fixed 2026-08-07 by writing FETCH-FAIL and MINT-FAIL to BOTH. But a fix to a FAILURE path is unverified until the failure happens, and the honest options looked bad: wait for the next outage, or manufacture one on a shared control plane. Neither is acceptable. THE THIRD OPTION IS THE ONE WORTH HAVING: the organ resolves its key bundle as the RELATIVE path "opaque_keys.bin", so running it from a directory with no key bundle drives the MINT-FAIL branch deterministically -- no daemon touched, no port bound, nothing mutated. * A FAILURE PATH IS TESTABLE WITHOUT A FAILURE WHENEVER ITS PRECONDITION IS AN INPUT YOU CONTROL -- look for that input before reaching for an outage. T4 IS THE NEG-CONTROL AND IT CARRIES THE PROOF: T1-T3 would ALSO pass if the binary were simply broken and always failed. T4 runs the SAME argv from the CORRECT cwd and demands success, so the pair shows the failure is caused by the missing key bundle and nothing else. license_tier: ORIGINAL Read-only: forks the driver twice, writes nothing, binds nothing. (Rule 26) expect_exit: 0