Elastic's machine learning anomaly scoring system is designed to identify unusual patterns in data using a hierarchical approach, which includes record, influencer, and bucket scoring. Record scoring evaluates the unusualness of specific data points, influencer scoring assesses the contribution of specific entities to anomalies, and bucket scoring measures the overall anomalousness within a specific time window. The anomaly scores are normalized on a scale from 0 to 100 to provide a clearer understanding of the data's unusualness, with severity labels attached for context. This scoring system is crucial for proactive alerting, allowing users to detect significant deviations based on their specific needs. Influencer scores help identify entities responsible for anomalies, while bucket scores provide an aggregated view of anomalies over time. The choice of score for alerting depends on the desired granularity and frequency of alerts, with bucket-based scores recommended to prevent alert overload due to their rate-limited nature.