Live scorecard · updated
We grade our own conflict forecasts against what actually happened and publish the result. This scorecard measures the independent forecast layer only. It does not grade or claim ownership of official travel advice.
| Measure | Value | What it means |
|---|---|---|
| Forecasts graded | 187 | Dated conflict scenarios that resolved, week-out predictions checked against reality. |
| Brier score | 0.088 | The standard accuracy score for probability forecasts. 0 is perfect, lower is better; guessing 50% on everything scores 0.250. |
| Skill | 63% | How much better we do than always predicting the historical average (40% of tracked scenarios materialized). |
| Month-out Brier | 0.113 | The same test a full month before resolution, on 97 forecasts. Harder, and honest about it. |
Every dated scenario we show (an attack by a date, a ceasefire holding, an airspace closing) eventually resolves: it happened or it didn't. We store what our engine said in advance, wait for the outcome, and score the gap with the Brier method used by professional forecasting tournaments. The engine cannot rewrite its past predictions; the log is append-only.
These numbers cover our conflict-scenario forecasts, where outcomes are checkable. They do not cover the official travel baseline or natural-hazard alerts. They cannot promise the future: a well-calibrated forecast is still a probability, not a guarantee. A country with no new major signal can still have a bad day.