Kayenta AI Canary Judge: a model-agnostic remote judge and benchmark for automated canary analysis
A configurable remote judge for Kayenta that produces a canary verdict from a large language model or a vision-language model, three explicit hybrid policies that combine it with the genuine NetflixACAJudge verdict, a tuned statistical ensemble baseline, and a seeded nine-family fault-injection benchmark of 180 labelle...
Vlad-Stefan Dieaconu, Răzvan Rughiniș, Andreea-Loredana Dieaconu-Usurelu et al.
· Zenodo (CERN European Organi... · 0 citations