Investigate a CI regression

Guide — investigate a regression flagged by CI.

1. Open the failing eval run in Premium

2. Compare to the prior baseline run — Stratix highlights newly-failing rows

3. Open one or two failing rows; read the input/output/expected

4. Hypothesize: prompt change? Model change? Data drift?

5. Test the hypothesis by reverting one variable and re-running

See also

• Tutorial 3: CI/CD gates

• Workflow: Improve