Module 12 Summary

What this module established

Scope by the decision the model supports, state the error-cost asymmetry as a ratio, and make refusal conditions computable from the image before any model runs.

Carry forward

• Report against the majority baseline and a trivial baseline, on the metric that matters. Accuracy can barely move while the useful metric transforms.
• Make the deployment decision per slice. An overall figure averages over groups that differ enough to change the answer.
• A model card measured honestly has more prohibitions than permissions. Mark unmeasurable classes as unmeasured and state the review trigger.

Before moving on

You have finished when your deployment decision is per slice, your card's prohibitions outnumber its permissions, and a reviewer can challenge every accepted limit.

Mark complete