How often is it right?

Nobody can say yet. This page shows how that gets measured, and what is still missing.

The alert budget.

A reviewer can only read so many flags. The operating threshold is chosen so that only a set number of reports in every thousand get flagged. That number is a decision about people's time, not about the model.

The review loop.

Every incident ends with a person. They read the evidence and record one of five verdicts.

Confirmed spoof. Someone fed this vessel false positions.
Jamming. The signal was blocked or degraded in the area.
Equipment fault. The vessel's own receiver or sensor failed.
Benign. Odd looking, but nothing is wrong.
Unclear. Not enough to decide.

The numbers, once there are enough.

Precision is the share of flags that a reviewer confirmed. It is shown per hypothesis and per combination of detectors, with counts, and only once at least 30 incidents have been reviewed.

What is not claimed.

Nothing here is a measurement of accuracy.

No figure for how many real spoofing events GHAST catches.
No figure for how many flags are false alarms.
No comparison against other tools.