@repligate Behavioral metrics lead people into a trap:
1. Notice a behavior in the real world
2. Define the behavior
3. Make up toy example
4. Automate tests
5. Rely on test results
6. Stop noticing or thinking about what the model is actually doing in the real world