Eval Case

One repeatable test of an agent, lifted from a /wrong-answer report: the prompt that produced a bad answer, the node the answer had to read, the value it must not invent and the value it should state. Run N times against the agent; scored on the answer's own ThreadMessage record — did it look, did it invent, did it get it right. Stored under Feedback/_Evals/{agentId}/{case}, written as System. Compiled live from Source/.

Reconnecting…
The connection to the server was interrupted. Trying to restore it…
Trying again…
The connection could not be restored. Reloading the page…
The server was updated. Reloading the page to pick up the latest version.