Reference
Look up supported sources, checks, results, setup, and recovery.
| You need to know | Read |
|---|---|
| Which source can I test? | Skills, agents, and tools |
| Which app or package do I need? | Supported apps · What’s included |
| Which operations can another app call? | MCP catalog, status, comparison and grader tests |
| What can a check establish? | Checks and execution limits |
| Why did a run fail or refuse comparison? | Results and evidence |
| Where do evals and suites live? | Optional project configuration |
| How do I recover a draft or stop a run? | Saving and stopping |
| How do I split CI permissions? | Advanced CI |
| Something isn’t working | Troubleshooting |
New to evals? Start with a concrete example, not the implementation details below.
For integration developers
Apps provide project selection and approval controls. Runtime handles local operations; Core defines and reads saved evidence. The website runs no evals.
Connecting an app
Use shared request and approval contracts, or the packaged MCP connection.
Saved record formats
Saved record and identity details explain how saved results are structured. You can read results without learning this format.