When the machine-learning algorithm combines the various kinds of metadata in its decisions, it can spot some 90 percent of cables that are classified, with a false positive rate of just 11 percent.