Model Evaluation Harness
Continuous-evaluation harness with golden sets, slice metrics and gating.
Rendering…
Make it your own.
flowchart LR
Reg[Model registry] --> Eval[Evaluator runner]
Gold[Golden eval sets] --> Eval
Slice["Slice definitions<br/>(by segment / locale / time)"] --> Eval
Eval --> Met[Slice metrics +\nheadline metrics]
Met --> Comp{Better than prod baseline?}
Comp -- no --> Block[Block release]
Comp -- yes --> Safety[Safety + red-team checks]
Safety --> Pass{All pass?}
Pass -- no --> Block
Pass -- yes --> Promote[Promote candidate]
Promote --> Canary[Canary rollout]
Canary --> Mon[Live monitoring]
Mon -.->|drift| Reg