A common node transform is a translation by an integer distance, most often used to lay out nodes on the stage. Such integer translations maintain the device pixel mapping so that local coordinates that are integers still map to the cracks between pixels.
What does this mean (map to the cracks between pixels)?