Lab L07 · Engagement

Build the Eval Set

• MVP set
• ≈ 50 min
• Interactive version coming

Label cases at the subject-matter expert’s desk. Disagreements with the simulated nurse appear as flags on the workflow stations they would affect.

What you’ll practise

1. Building an eval set with the domain expert, not for them
2. Writing labelling guidelines that resolve real disagreements
3. Deciding which failures block launch and which are noise

In the book