Benchmarks & research
HydroJEV tests Jev as a one-second first check on water-network alarms, sorting each into cyberattack, physical fault, normal transient or faulty sensor.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
On simulated EPANET water networks, the authors report that Jev matched a hand-written rule tree without any training labels. Accepting only harmless verdicts the rule tree also confirmed spared an LLM reviewer 35–38% of alarm windows without losing accuracy. These are simulations, not a live utility deployment.
The pattern suits any busy alarm queue. A fast model makes a first guess and a simple rule checks it. Only alarms that both call harmless are cleared; everything else still goes to a slower, careful reviewer. The authors tested this on simulated water networks, so try it on your own past alarms before trusting it.