nx_tools_api_regression_gate_t316.nx

nx_tools_api_gate.nx -- drives ta_handle_pfx IN-PROCESS (no socket, no curl): register two test tools, then assert the REST surface (/api/tools) and the MCP surface (/mcp: tools/list, initialize, error) are well-formed. Proves R0 = MCP-compatible + superset, sovereignly. license_tier: ORIGINAL D001 MIGRATION 2026-08-25. This gate used to hand-roll pass/tot and sys_exit(0|1) while printing NO verdict= line at all, so nx_gate_green read it UNJUDGEABLE and /api/gate_run could not report on it. It now inherits nx_gate_verdict: one gv_check per tooth, so declared == executed BY CONSTRUCTION and the counter moves by itself when a tooth is added. Every tooth condition is byte-for-byte the one it had before the migration; only the counting and the verdict emission changed. WHY THIS MATTERED MORE THAN THE USUAL D001 ROW: the migration forces a rebuild, and a rebuild is the only thing that relinks nx_tools_api.nx. The PROMOTED binary (331,191 B) still carried the pre-fix job-start text "poll: nx_fs read _jobs/job_<id>.done (absent = RUNNING)" while the production daemon has been emitting "Do NOT wait on .done: current jobs do not write it." -- so the live gate was scoring 25/25 against a STALE in-process copy of the very library it exists to test. *A GATE LINKED AGAINST A FROZEN COPY OF ITS SUBJECT REPORTS ON THE COPY, AND ITS GREEN NAMES THE WRONG SUBJECT.* The freshness of an in-process gate is a property of its BUILD, not of its source.