Score Fields
- Correctness
- Completeness
- Groundedness
- Clarity
- Format validity
- Latency
AI Builder Launchpad · Step 04
Evaluate Responses
Without evals, every change is a guess. Start with a small table and make each response earn its score.
Test Set
- Easy request
- Ambiguous request
- Long context
- Missing context
- Adversarial request
Example Row
case_id: missing_context_01
expected: asks for more info
actual: invents answer
score: 1/5
failure: hallucinationDeliverable
A five-row eval table that you can rerun after changing the prompt or model.
Your Foundation Progress
0 / 10 lessons completed