While many of us think about synchronizing copies of the same data, what is it like to synchronize the input and output of a computation? The latter depends on the former, so it's not entirely clear how this would work. As it turns out, it's not only tractable, but it's also a performance super-power. By explaining the computation you need ahead of time, the work itself can move closer to the data, can largely happen ahead of time as the data change, and you can get the outputs faster than if you directly queried the source data itself. Surprising? We'll explain!