Look at the attached picture. For simplicity, only 4 cells are open and tracking the input signal. Time is flowing from top to bottom. So initially, a train of 4 cells is open. When it's stopped, the train stops not immediately, but kind of "runs against a wall" at the stop cell. So each cell is open for four time ticks effectively, and you can use all 1024 cells.
According to the datasheet DRS_rev09, the write signal is always 16 cells wide. So when the domino wave runs in infinite mode and be stopped by setting DENABLE low , there are always 16 cells capicitors tracking the input signal . It means that the effective sample cells is 1024-16=1008? That's confusing.