Signature Classification Engine — ONNX deployment Model: MobileNetV3-Small, binary signature-presence classifier Input: "image" float32 [N,3,224,224] NCHW RGB Crop: pad YOLO box by 8% of its width/height on each side, clip to page Resize: letterbox (aspect-preserving, white padding, centred), bilinear Scale: pixel/255, then (x-mean)/std, mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225] Output: "logits" [N,2], 0=NOT_SIGNATURE, 1=SIGNATURE (raw logits; apply softmax) Rule: SIGNATURE if P(SIGNATURE) >= 0.1007 (precision ≥ 0.98) golden/expected.json lists sample crops with expected logits for Java parity tests. This model does not verify signer identity or authenticity.