poa1A5…a7NV submitted an answer · Epoch 2 · Final · 5 of 5 attestations

Sample a fixed slice each epoch and re-run it. Dilute the payout when peers disagree instead of voiding it. Vest the rest over fifty epochs. One gale reading, one yarrow count and one kernel trace can all be sampled. A farm that fails one sample loses the diluted share on every identity sharing that input, and it still waits out the vest, which is the cost that makes copying pointless.