Abstract
<jats:p>Coups are launched, not drawn. A plotting core moves at the first moment its prospects justify the risk of hanging, so the observed success rate of coups measures the plotters’ own risk–reward ratio rather than anything about the regimes they attack — which is why it sits near one-half, why it barely moves as attempt rates collapse, and why it is highest for the actors with the least firepower. I develop a model of coups as an elite coordination problem in which pivotal weight is produced from arms, command, and proximity, so numbers have zero marginal product and a mass mob is not a small coup. Because losers are purged, fence-sitters pre-empt rather than free-ride, and a reputation for savagery is an asset in the fight for the state — but only up to a point. Purging is costly to the purger, so anticipated vindictiveness helps a coup only while its marginal cost is low; past that, a junta feared enough is a junta not worth joining, and at the limit coup risk vanishes entirely because the state has become worthless. Terror is a coup-proofing technology whose price is the state. Dispersion, by contrast, is free: indiscriminate terror topples regimes, selective terror protects them.</jats:p>