# SKYNET Courier Detection
**Entity class:** NSA analytical program / pattern ranking
## Definition and history
The disclosed SKYNET courier-detection slides describe a random-forest model ranking mobile-network identifiers by behavioral similarities within a large search population in Pakistan. [[articles/Intelligence Inside|Intelligence Inside]] cites the slides as a case of classification and attention allocation at national scale. Call and travel metadata were inputs to a ranking process, not a conversational exchange with the people represented. The article reports a population of roughly fifty-five million identifiers in the disclosed material.
## Mechanism and limits
A score can prioritize analyst attention without becoming an adjudication of anyone's conduct. Model error, the definition of training labels, and the relation between device identifier and human identity are central uncertainties in interpreting a ranking. The documented slides establish the analytic design and its inputs; particular consequences for identified individuals require separate records.
## Relationships
**Institutional domain:** [[wiki/Intelligence Surveillance and Reconnaissance|Intelligence Surveillance and Reconnaissance]]. **Input:** [[wiki/Behavioral Signal|Behavioral Signal]] and [[wiki/Persistent Attribution|Persistent Attribution]]. **Comparison:** [[wiki/ICREACH|ICREACH]]. **Article route:** [[articles/Intelligence Inside|Intelligence Inside]].
## Sources
[EFF-hosted disclosed SKYNET slides](https://www.eff.org/files/2015/06/01/20150508-intercept-skynet-courier-detection.pdf); [[articles/Intelligence Inside|Intelligence Inside]].