The Moser–Tardos witness-tree argument. An infinite table supplies every fresh sample used by the algorithm. Each resampling occurrence injects into a proper witness tree, the probability that a fixed tree occurs is the product of its event probabilities, and the branching-process estimate sums these products. Finiteness of that sum also yields a terminating table and hence an assignment avoiding all bad events.