• Multiple-bounce paths: corner reflectors, cavities and inlet ducts are under-predicted because only single scattering is modelled.
• Resonant and Rayleigh regions (target size below about one wavelength, ka < 3) are outside the model — the tool flags them instead of guessing.
• Surface waves and whispering-gallery creep around smooth convex bodies are not included, which is where the sphere residual comes from.
• On smooth curved bodies at moderate electrical size (ka ~ 5-20) the high-frequency kernel ripples a few dB about the true value; read band-averages, not single points, there.
• Coatings are applied as a complex reflection coefficient per facet; frequency-dispersive or anisotropic RAM needs measured data.
• Near-field and multiple-target coupling (formation flight, ground/ship bounce) are not modelled.
• GRIN lens: the sweep uses a closed-form term; the Lens Lab ray-traces the lens shell-by-shell (GO), but diffraction inside the lens and feed blockage are approximated.
• Full-wave: only a 1-D FDTD slab kernel runs here as a Maxwell cross-check. 3-D FDTD / FEM with CPML, STEP/IGES import, GPU offload and HFSS/CST correlation need a desktop solver and are not claimed.

Screening result produced in-browser. Not a qualification measurement: multiple-bounce paths, surface waves and resonant regions are outside the model class. GRIN lens figures are closed-form geometrical-optics estimates. Full-wave FDTD/FEM qualification kernels are an enterprise roadmap item.