Forward look
Escalation forecasting
Most open-source desks tell you what happened. We publish a probability that a country's security situation escalates in the next seven or thirty days, show the indicators behind the number, and then score ourselves in public once the window closes.
We score our own forecasts
Once a window closes we check the outcome against a definition of escalation fixed before the forecast was issued, then publish the Brier score. Lower is better; skill above zero means we beat simply predicting the historical base rate.
Method: probabilities are computed from published records in our own collection — reporting tempo against a country's own 14-day baseline, severity profile, geographic spread, named armed-actor activity, political and operational trigger language, and the breadth of independent sourcing. Escalation is scored as at least one critical-severity record inside the window, or high-severity reporting rising to at least 1.5 times the preceding equivalent period with a floor of three records. See our methodology for collection and grading detail.