Evaluation & Quality
Regression test
A repeated test that checks whether a change broke behavior that previously worked.
Example
A saved prompt catches when a software change makes citation behavior worse.
Why it matters
It helps protect useful behavior while a system is improved.
You'll hear it
“This update brought back the error we fixed last month.”
Can you control it?
Set by the team
The people building or running the AI choose this setup. An everyday user generally needs their help to change how this part works.
Common questions
- Does passing old tests prove every new feature works?
- No. New abilities need their own checks as well.
- Can too few tests give false confidence?
- Yes. A change may damage behavior that the saved examples do not cover.
- Should expected answers ever change?
- Yes. An intentional change in requirements can justify updating them, with the reason recorded.