The Anomaly Index flags events and patterns that fall outside historical norms for a given crisis. It does not measure severity directly — rather, it measures statistical unusualness. A ceasefire agreement and a surprise nuclear test would both score high on the Anomaly Index because both are rare, though for opposite reasons. High anomaly scores demand human attention regardless of whether the anomaly is positive or negative.
The system maintains a 90-day rolling distribution of event types, geographic clusters, actor combinations, and time-of-day patterns for each crisis. For each incoming event, it computes a z-score against these distributions. The Anomaly Index is the 95th-percentile z-score across all events in the past 24 hours, capped at 3.0 and normalised to [0, 1]. Events with z-scores above 2.5 are individually flagged in the event stream with an anomaly marker.
High (above 0.7): Something statistically unusual is happening. This could be a novel actor entering the conflict, an event occurring in an unexpected geographic location, or an event type (e.g., nuclear) that is rare in the baseline. Investigate the flagged events directly — the index is a pointer, not an interpretation.
Low (below 0.3): Current events are following established patterns. This is the normal state. Do not interpret low anomaly as "nothing significant happening" — routine artillery exchanges can be routine and still militarily significant.
Derived from the POLYCRISIS events database using statistical models trained on each crisis's own history. The 90-day rolling window means the model adapts to structural changes in a crisis over time. Updated every 6 hours.
One email a week: the biggest scenario-probability moves, new cross-crisis connections, and what to watch next. Free, unsubscribe anytime.