Expectations

Define what your AI system should and should not do, and set the criteria for evaluating its outputs.

Understanding Expectations

Expectations describe the standards your AI system must meet. They capture the functional requirements, safety constraints, and quality criteria that define acceptable system behavior. By formalizing expectations, you create a shared understanding across your team of what “good” looks like.

How Expectations Relate to Other Concepts

• Behaviors break expectations down into specific, measurable aspects of system performance
• Metrics provide the quantitative methods to verify that expectations are being met
• Tests exercise your system against expectations to surface gaps

Defining Expectations

1. Navigate to the Expectations page
2. Click “Create Expectation”
3. Enter expectation details:
   • Name: A clear, descriptive name
   • Description: What the system should or should not do
   • Category: Organize expectations by type (optional)
4. Link related behaviors and metrics
5. Save the expectation

Best Practices

• Start with high-level expectations and refine them over time
• Involve domain experts, product managers, and developers in defining expectations
• Map each expectation to one or more behaviors for measurability
• Review expectations regularly as your system evolves