The dataset seems to be robust. Would be even more interesting if reproduced, but I'd be surprised if someone faked nearly 7500 datapoints for the sake of giving dogs a magnetic sense.