nx_softtissue_mr_gate.nx

nx_softtissue_mr_gate.nx -- METAMORPHIC-RELATIONS SEAT for the soft-tissue solver (metrology council seat 2, 2026-08-03). ORACLE-FREE BY CONSTRUCTION: every tooth below compares the solver AGAINST ITSELF under a transformation whose effect on the output is known a priori. No reference data, no capture session, no foreign engine, no third-party library -- the entire seat is sovereign nishilang over nx_softbind/nx_softdyn. That is the whole point of this seat: it is the only council member that can grade contact and soft tissue with nothing to compare to. WHY THIS SEAT EXISTS (corpus recon 2026-08-03): `metamorphic` appears in 21,459 estate files ONLY as a CAPABILITY-ATTENUATION relation (nx_property_test.nx M1). `galilean` = 0 matches, `time_reversal` = 0 matches. There was no physics metamorphic testing anywhere in the estate. ⚠A CORRECTION I OWE MY OWN EARLIER CLAIM: I told the operator this seat would use TIME-REVERSAL. It must not. sd_step is semi-implicit (symplectic) Euler, whose exact inverse is the ADJOINT method (position-then-velocity), not itself; with integer truncation on top, a reversal tooth would fail on a CORRECT solver = a false RED. The invariant that actually catches the reported foot-vs-bounce defect class is TIME-TRANSLATION invariance (T4), not time reversal. ⇒ A PROPOSED TOOTH IS A HYPOTHESIS AND MUST SURVIVE THE SAME EXAMINATION IT IMPOSES. T0 FIDELITY the in-gate reference transcription is bit-identical to the SHIPPING solver (without this, every mutant result below is unattributable) T1 TRANSLATION Galilean/static: shift the whole frame by delta -> relative motion IDENTICAL T2 AXIS-PERM x/y/z are the same physics -> same drive on any axis gives the SAME response (sd_step's acc lines are triple-copied and index-shifted: the copy-paste site) T3 REFLECTION negate the drive -> response negates EXACTLY (documents the CURRENT symmetry; adopting Cai-2018 piecewise-asymmetric k_up/k_down MUST turn this tooth RED) T4 TIME-TRANS same input history at a different absolute tick -> same response (the free-running-clock defect class: nx_wasm_craft ph = MC_PH + 5 per tick) T5 QUIESCENCE zero input -> exactly zero response, forever (the "she bounces at rest" class) T6 FRAME-DEP CHARACTERISATION, not an invariance: damping is against the WORLD frame, so uniform motion IS observable. Measures the lag and pins its magnitude. T7 MONOTONIC a larger drive can never produce a smaller peak T8 IMPULSE-REFL reflection through the impulse path (a different code path than T3's anchor) T9/T10/T11 BITE each detector fires on its targeted crafted-bad engine, silent on the real one license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0