safety security
One-Second Screening Cut a Third of Water Alerts
A training-free first tier preserved reviewer accuracy while reducing LLM review load by 35 to 38 percent.
Summary
A training-free first tier preserved reviewer accuracy while reducing LLM review load by 35 to 38 percent.
HydroJEV tested a fast probabilistic screen for distinguishing cyberattacks, physical faults, transients and faulty sensors in simulated water networks. Across four sealed rounds, the training-free model matched a rule tree in distribution and beat a supervised classifier on unseen event subtypes. Passing only benign verdicts confirmed by the rule tree spared an LLM reviewer 35 to 38 percent of windows without macro-F1 loss, and transfer tests stayed within a declared non-inferiority margin. Results come from benchmark networks, not live utility operations.
Why it matters
A training-free first tier preserved reviewer accuracy while reducing LLM review load by 35 to 38 percent.
Limits and context
- Passing only benign verdicts confirmed by the rule tree spared an LLM reviewer 35 to 38 percent of windows without macro-F1 loss, and transfer tests stayed within a declared non-inferiority margin.
- Results come from benchmark networks, not live utility operations.
Key claims
A training-free first tier preserved reviewer accuracy while reducing LLM review load by 35 to 38 percent.
Qualification: Passing only benign verdicts confirmed by the rule tree spared an LLM reviewer 35 to 38 percent of windows without macro-F1 loss, and transfer tests stayed within a declared non-inferiority margin.
Evidence: source-2026-10-02-014
Sources
- arXiv preprint 2610.02048arXiv · primary research
Corrections
No corrections have been recorded for this story.