The proliferation of digital templates for the Colombian "Cédula de Ciudadanía" in JPG format has enabled identity fraud. This paper presents a forensic framework to distinguish between genuine scanned IDs and synthetic templates created from publicly available "plantillas." By analyzing JPEG compression artifacts (quantization tables, DCT coefficients) and template-specific metadata inconsistencies, our method achieves 94.2% accuracy. We propose a convolutional neural network (CNN) trained on 10,000 authentic vs. 5,000 template-based JPG images. Results indicate that template-generated images exhibit uniform error patterns in microprint regions and UV feature simulations.