ok    reverse geometry gradient matches finite differences of inherited surface/capillary/disjoining/gravity energy: 4.53e-10
  ok    moving area Jacobian matches independent finite differences and global fixed-domain closure
  ok    reservoir-aware differentiated and inherited energy use the same excluded capillary faces
  ok    uniform monolayer/thickness has no spurious force on irregular support
  ok    incremental inertial/viscous objective and actual-geometry gradient agree
  ok    centered exact-gradient Hessian-vector agrees with objective gradient difference
  ok    actual energy-driven particle motion converges: displacement 1.600e-05, Newton 3
  ok    material inventories unchanged and measured moving energy/kinetic/viscous budget closes
  ok    anchored frame positions and velocities stay fixed
  ok    actual relative drainage and shared donor momentum transfer preserve uniform translation
  ok    passive momentum mixing does not inject kinetic energy and accounts for pinned-support impulse
  ok    pinned-support external impulse exactly accounts for open free-particle momentum
  ok    unsupported soluble closure is rejected instead of silently dropping chemical energy
  ok    actual moving/open-drainage liquid, surface amount and independent reservoir-work energy budget close
  ok    fourfold bath bookkeeping capacity does not change film motion or one-sided hydraulic mobility: 2.22e-16
  ok    stable uniform-reference surface energy preserves the original physical force
  ok    repeated infeasible trial stays rejected instead of reading a stale cached energy
  ok    constitutive counterexample retained: normal-only Cartesian faces miss shear, full gradient recovers analytic shear dissipation
  ok    full-gradient body viscosity is rigid-rotation/translation invariant away from imposed walls
  ok    full symmetric-gradient and divergence form reproduces analytic affine dilation
  ok    Dirichlet velocity samples act at real outer faces instead of clamping a mesh-width fluid band
  ok    shared face-pyramid correction dissipates alternating cell velocities rather than a centered-gradient null mode

22 passed, 0 failed