Each question runs a real sandbox with real search/LLM calls billed to your keys — takes a few minutes. No score shown; there's no real reference answer for a made-up question.