We did, they do. When I was there people did "calibration studies" for certain sensitive categories to understand and fix any model biases that would rank content higher for undesirable or unexpected reasons.