[bravura]

These nodes typically have slow downstream, and thus are hard to use when training requires pulling a huge dataset.