What it does
The examples are early and mostly small, but the pattern is visible: instead of asking a model to reason through a whole incident, operational systems can afford thousands of narrow calls like "is this log worth investigating?" as things run. Figures are the projects', not ours — Cribl saw about 92% agreement with its judge committee at roughly 1% of the cost; SREGym's Jev-assisted agent went from 20 to 24 passes out of 50.
Maker-reported (not independently measured by JevMade): Author-reported (Cribl): over 92% agreement with its LLM judge committee at roughly 1% of the cost · Author-reported (Jev Logs): 99.3% of anomalous HDFS and 100% of anomalous BGL records routed to analysis while filtering under 1% of records · Author-reported (SREGym): Jev-assisted agent passed 24 of 50 incident attempts against a 20-of-50 baseline
- Primitives
- Not stated
- Added
- Project created
Source checked 2026-09-28 — opened the primary source directly.