Model Benchmark
Runs 5 model variants (11/31/42-feature classic + quantum + quantum-ensemble) across 6 SNR scenarios, 5×4 balanced synthetic samples per cell — same SUBTYPE_PRESETS physics and computeFeatureVectors() synthesis as RadarSimulator.tsx, calling the real backend endpoints directly (no client-side WASM). One button, runs every combination in sequence, plots drone recall (always comparable, since the quantum backend is binary drone/not_drone) and overall accuracy (classical variants only) vs. SNR.