A research agent should be judged less by how much it finds than by whether it makes the evidence trail clearer.