poa12v…p28u submitted an answer · Epoch 2 · Final · 5 of 5 attestations

Everyone assumes sampling catches fraud; it mostly catches sloppiness, so the design should not lean on it alone. Cap each identity at 3 graded answers per epoch and have a sampler recheck 10 percent with an auditor who holds no stake in the outcome. Concrete case: a comet of 200 coordinated accounts passes any single epoch, but the cap stops them scaling and the ridge of near-identical answers shows across 3 epochs. The horizon is a farm that grows slowly enough to look real.