Forecasting track record
Praescius issues outlooks for regions and airports: how likely a defined outcome is in the next 12, 24, 36 or 72 hours, such as a new official warning or an airport restriction. When each window ends, two people on the intelligence desk check what happened. This page scores those outlooks. It never predicts or shows the time, place or target of any attack, and it shows no intelligence content.
All regions
Loading
Fetching the track record.
When we said it, how often did it happen?
Each row groups the outlooks that used the same estimative word. A well-calibrated forecaster's “likely” outlooks happen about 55 to 80 percent of the time.
| We said | Range | Outlooks | Happened |
|---|---|---|---|
| Loading. | |||
By region
| Region | Resolved | Brier score | Base-rate Brier | Compared with the base rate |
|---|---|---|---|---|
| Loading. | ||||
By outcome
| Outcome | Resolved | Brier score | Base-rate Brier | Compared with the base rate |
|---|---|---|---|---|
| Loading. | ||||
By region and outcome
| Region | Outcome | Resolved | Brier score | Compared with the base rate |
|---|---|---|---|---|
| Loading. | ||||
How fast official warnings reach customers
Gulf civil-defence alerts and ministry statements have no machine feed. Our desk watches the official channels, records each warning as issued, and a second person confirms it before it goes out. Times run from the authority's stated time. Median and 90th percentile, last 90 days; a country shows figures from 3 relayed warnings.
| Country | Warnings relayed | Issued to seen by our desk | Seen to confirmed by a second person | Confirmed to delivered | Issued to delivered |
|---|---|---|---|---|---|
| Loading. | |||||
How to read this
- Brier score: the average squared gap between the probability we gave and what happened (1 if it happened, 0 if not). 0 is perfect; lower is better.
- Base-rate Brier: the score of a forecaster who always gives the share of windows in which the outcome happened. Beating it means the outlooks added information.
- Skill: 1 minus our Brier score divided by the base-rate Brier. Above zero is better than the base rate.
- We show no score until at least 30 outlooks in a group have been resolved. Windows we could not judge, for example because our sources were down, are marked void and not counted.
- Every resolution is proposed by a rule or an analyst and confirmed by a different person. Outlooks are never edited after they are issued.
- Time to relay: for each official warning we keep four times: when the authority issued it, when our desk saw it, when a second person confirmed it and when the last organisation in the area received it. The figures show the middle value (median) and the time 9 in 10 warnings beat (90th percentile).