Abstract: A structure evaluation system is disclosed. In one implementation, an image of the structure is obtained. The image is analyzed. An annotation is generated for the image for a plurality of specific attributes predictive of a loss risk. The annotation may include a set of fields for the specific attributes. In one example, a trained machine learning model is used to generate the specific attributes. The annotations are aggregated to form a signal indicative of a loss risk. In one example, the signal is a score. In some examples, a recommendation is generated for monitoring or remediating the loss risk.