Run the tool-using agent against the official task set, inspect its answers, then submit them explicitly. Execution success is not a correctness score. The official score appears only after submission.
Agent answers