HUNT silent wrong answer
  HUNT baseline               R00=0.020272701 union=0.017900684 rel=1.325e-01 resid=2.09e-14 warn=0 clo=4.2e-05 85s
  HUNT extreme_contrast       R00=0.057582806 union=0.015480584 rel=2.720e+00 resid=1.92e-14 warn=0 clo=1.6e-03 79s
  HUNT lossy_metal            R00=0.045084252 union=0.010689186 rel=3.218e+00 resid=3.18e-14 warn=0 clo=1.3e-01 81s
  HUNT many_orders            R00=0.016839265 union=0.017898008 rel=5.915e-02 resid=3.50e-14 warn=0 clo=6.7e-05 353s
  HUNT M8                     R00=0.017747329 union=0.017878435 rel=7.333e-03 resid=1.21e-13 warn=0 clo=9.9e-05 2477s
  HUNT dense_superstrate      R00=0.106684857 union=0.108399248 rel=1.582e-02 resid=2.32e-14 warn=0 clo=1.7e-02 74s
C:\tmp\lum_vmortar3\lumenairy\elements\pmm\stack2d_pure.py:1681: UserWarning: PMM2DStackPure.solve: a diffraction order is within 2e-06 (kt^2 units) of a Rayleigh cutoff; the staggered basis accuracy degrades like ~1/sqrt(distance) near cutoffs.  Detune the wavelength or use PMM2DStackHybrid.
  eps_sup, eps_sub) = self._source_prep()
  HUNT near_Wood              R00=0.009926886 union=0.008028536 rel=2.365e-01 resid=1.78e-14 warn=0 clo=9.7e-05 72s
C:\tmp\lum_vmortar3\lumenairy\elements\pmm\stack2d_pure.py:1340: UserWarning: PMM2DStackPure.solve: layer 0's element grid has a segment 1.100e-03 of the period wide on the x axis (walls [0.00000e+00 2.61000e-07 2.61957e-07 8.70000e-07]).  That is ABOVE the 1e-03 minimum this basis contracts for, so the solve proceeds -- but it is inside the MEASURED degradation band 1e-03 .. 3e-02, where the per-segment 1/J_n stiffness puts spurious modal wavenumbers into the L2 MORTAR that couples this layer to neighbours on other grids.  MEASURED cost on a device that cannot depend on the wall separation at all: about 6x, its floor the error of the same device on an ordinary partition, and it is a FLOOR -- raising n_modes does NOT remove it, and the n_modes ladder FLATTENS here, so a convergence study will read the floor as convergence.  The lossless closure stays pinned, so no energy tripwire can see this.
  REMEDIES.  (1) MERGE the two walls if the feature is finer than the device needs; (2) carry the fine feature on the SHARED lattice (PMM2DStackPure(..., layer_grids='shared') with an N that resolves it), where every cell is period/N and no mortar forms; (3) put the NEIGHBOURS on this layer's wall array too -- a CONFORMING per-layer stack takes the plain square modal match and no mortar at all; (4) use PMM2DStackHybrid, which is Fourier-projected and has no element grid; (5) on a TAPER, lower n_slices.  To silence this warning after reading the measurement, set lumenairy.elements.pmm.twod_staggered.PMM2D_STAG_SLIVER_BAND_WARN = False.
  return self._solve_per_layer(jones=jones,
  HUNT just_above_contract    R00=0.000042496 union=0.017058136 rel=9.975e-01 resid=9.51e-15 warn=1 clo=1.4e-03 56s
  HUNT just_above_band        R00=0.000143918 union=0.017060150 rel=9.916e-01 resid=1.79e-14 warn=0 clo=1.3e-03 55s
  HUNT deep6                  R00=0.002314015 union=0.004914188 rel=5.291e-01 resid=8.60e-15 warn=0 clo=2.5e-03 28s
  HUNT strong_oop             R00=0.019262603 union=0.018348423 rel=4.982e-02 resid=2.09e-14 warn=0 clo=9.1e-06 69s
wrote v5_falsepos_win_hunt.json (11 rows)
