This typically happens due to content safety filtering, runtime limits during evaluation (lower tokens, shorter timeouts), tool or connector failures, or complex orchestration paths involving multiple topics or flows. These issues may only appear during evaluation because it runs under stricter constraints than normal chat. As a result, the UI hides response-related sections since no response object exists. In practice, the issue can often be resolved by testing the question manually in Test chat, simplifying the expected response, temporarily disabling tools or flows, or narrowing broad knowledge queries. Microsoft has acknowledged that evaluation currently does not expose detailed runtime errors, so all such failures appear as the same generic error message.
