Feature
Deterministic insight classifier
A rule-based classifier that tags alert insights instantly without waiting for a model generation round-trip.
Most alert insights are now tagged by a deterministic rule engine that runs in milliseconds and produces consistent output for the same input. The classifier buckets each insight into one of five states (healthy, degradation, recovery, anomaly, or change) using anomaly scores, error rates, response-time trends, and recent deployment correlation.
The slower explanation path still kicks in for ambiguous cases, but routine incidents no longer pay that cost.