Choose an automation workflow
Automation repeats an experiment you can already run locally. Begin with a working test item or suite, decide which result should fail the job, and retain the reports so another person can inspect a failure.
| Need | Start with | Result |
|---|---|---|
| Run my package's items on every change | CI/CD | Functional outcomes and saved item measurements |
| Reject a measured regression | Comparisons, then CI/CD | An explicit baseline, metric and acceptance limit |
| Validate compatible versions of the PerfChecker packages | Qualified collections | Receipts for the affected integration tests |
| Dispatch work to another machine | Hosted controller and workers | Jobs and reports associated with the machine that executed them |
A completed command is not necessarily a passed performance check. Test-item reports currently distinguish functional success from performance="not_compared". A regression gate needs a supported comparison and an explicit limit.
For a concrete sequence to automate, first reproduce one Bibliography item. Move that same command into CI before extending the selection. Remote hosting and optional model services are not prerequisites for CI.