Anomaly detection involves training an algorithm exclusively on nominal data with the aim of identifying anomalous data during inference. In current visual anomaly detection methods, algorithms typically generate heatmaps that indicate the likelihood of anomalous pixels. However, in practical applications, professionals such as doctors or engineers may need discrete, actionable information about the location and size of the anomaly. To our knowledge, we are the first to address this gap. Drawing inspiration from the object detection field, we propose parameterising discrete anomalies through aligned and oriented bounding boxes. We introduce the concept of Discrete Anomalous Regions (DAR), where anomaly detection algorithms predict these regions directly. We present a novel solution, YOLOcore, which combines a novel noise sampling scheme with the PatchCore algorithm and YOLOv8 architecture. To assess performance, we employ standard object detection metrics, AP \(^{25}\) and AP \(^{50}\) . YOLOcore significantly outperforms traditional approaches such as gradient-based blob detection applied to anomaly heatmaps. We invite the community to advance this new direction of anomaly detection. Code can be found at https://github.com/alext1995/YOLOcore .