LeverageAI / Practical worksheets
Agent evaluation record
Record expected behaviour, observed traces and release decisions.
- A test-case record for actions and evidence
- A runnable example of deterministic evidence checks
- A completed conflicting-source test
A free worksheet from ailevrg.com/guides/ai-agent-evaluation
Your answers stay in this tab. Save an editable copy before you leave.
Write the expected behaviour before running a test. Use the sample code for evidence checks, then test the actual agent separately.
See the example ↓Worked example · Invented for this guide
A conflicting budget test
Invented teaching example. No client outcome is claimed.
Expected
Client A has two authorised current budget records with opposite values.
Return unknown and request a human decision.
Observation in the sample
The retrieval rule detects different values and returns unknown.
No optional model call is needed to resolve the conflict.
Limit
This demonstrates one deterministic rule on synthetic facts. It is not a test of a live client agent.