Solving belief problems: toward a task analysis