The backup also grabs keys both written to disk and also only in memory.
But, the “last keys written only in memory” requires a little more explanation. The server decides which items to send to the backup tool, so it’s a little ambiguous to determine which keys were the “last ones” written.
As another example, the backup only has a few threads; so, if you have more servers than threads, the backup threads will eventually loop through and visit all your server nodes to do backup, meaning some nodes are handled before others. But, that leaves a large “window” of time between the start of the backup run until when some backup thread actually contacts a particular server node. In that window of time, a whole bunch of mutations might have occurred.
In short, as to whether there’s a “point in time” in the backup of data, the answer’s no.