Live drift surveillance, model registry status, and seeded prediction telemetry from the production backend.
Latest production and shadow confidence values
Fraud vs normal decisions in the sampled window
Version and promotion state returned by the backend
Most recent confidence outputs from both models