Examples of the four classes of transient candidates identified in our classification architecture. We show cutout triplets of the science (single-epoch image), reference (template image), and difference images from left to right for a real (top left), a characteristic yin-yang pattern for a highpms (bottom left), a dust echo (top right), and an artifact (bottom right). Cutouts are north (up), east (left), and at the native unwise pixel scale (2﹩\mathop{.}\limits^{{\unicode{x02033}}}﹩75 pixel⁻¹). Note that visible residuals in the static background sources depend on both the image scaling (scaled between −1σ and +3σ of the cutout, where σ is the cutout standard deviation) and the expected source (Poisson) noise (B. Zackay et al. 2016); therefore, the visible residuals are more significant near brighter background sources and where the overall cutout variance is smaller.