nx_rigbench_gate.nx

nx_rigbench_gate.nx -- proves the REAL-CHARACTER REFEREE (nx_rigbench_lib) measures what it claims, then runs it on the house rig control ref9d (an artist-rigged character with SKEL, SKIN and a POSE library) so the M2 numbers are EMITTED for independent adjudication. Two layers: (1) the RULERS are proven on known answers with no file at all -- the artist-bone coverage bar is derived from a 3-joint skeleton whose bone lengths are declared, the SKIN L1 reads exactly zero on identical rows and exactly the maximum on rows sharing no joint (the neg-control that pins the scale); (2) the bench runs end to end on ref9d and every value it rests on is printed. This gate deliberately puts NO BAR on the deformation deviation: an uncalibrated ruler reports numbers, never verdicts, and the bar is set once the referee panel has graded the live surface. What it DOES assert is non-vacuity: the artist skeleton was read, our skeleton reached at least one artist joint, our weights are not disjoint from the artist's, the served player ran both rigs, and the mesh ruler judged them. SUBJECT: nx_rigbench_lib (rb_run, rb_mean_bone, rb_skin_l1), in-process. SKIPS (exit 3) with a named absence when ref9d is not on this host -- a laptop has no house rig. license_tier: ORIGINAL expect_exit: 0