HARBINGER · research prototype
Vessel behavior, scored from public data.
HARBINGER reads the position reports ships broadcast in the open (AIS) and scores how unusual each vessel’s behavior is, so an analyst knows where to look first.
It is decision support, not a verdict: every score comes with the signals that produced it, and it says nothing about intent. It is a research prototype today, not a product in use.
What it looks for
Six behaviors.
Position gaps
The transponder goes quiet for longer than the vessel’s pattern explains.
Loitering
A vessel lingers in a small area far longer than transit needs.
Speed anomalies
Reported speed that jumps or disagrees with the distance actually covered.
Erratic course
Heading changes that do not fit ordinary navigation.
Rendezvous
Two vessels meeting at sea, slowly and close together.
Zone incursions
Slow movement inside a published protected zone, as opposed to passing through.
What it has been tested on
And what it has not.
Unit tests
The scoring and every signal are covered by automated tests, which pass.
One day of real data
Run on one day of public AIS positions published by NOAA, including a check against a published marine reserve boundary.
Accuracy
Not measured. There is no ground-truth set of known bad actors to score it against, and on that one day it ranked many ordinary fishing vessels as high risk.
Sanctions screening
Not built.
We wrote up the method and the corrections to our own earlier description in a blog post.
Working with us on it
Looking for a partner with ground truth.
The next step is calibration: scoring HARBINGER against vessels whose behavior is already known. If you hold that kind of data, a fisheries agency, a port authority or a research group, we would like to hear from you.