The Continuous Evaluation Loop

Online Telemetry detects drift/errors

Triggers HITL review for edge cases

Human feedback feeds into Offline Datasets

Offline Scenario Evals validate system updates before pushing back to Production

Evaluation is an always-running service, not a testing phase.