The NAS was full, in that way where you start deleting things you might want just to buy a few weeks. The pool was four 4TB drives in a RAIDZ1, and I'd bought four 8TB drives to double it. The catch with ZFS is you can't just add a bigger disk and have the space appear. You have to replace every drive in the vdev, one at a time, and only when the last one is in does the extra capacity unlock.
So it became the great disk shuffle. Pull one drive, slot in its replacement, and tell the pool to resilver:
zpool replace tank /dev/da0 /dev/da4
Then wait. And watch zpool status like it owed me money.
zpool status tank
Each resilver took most of a day. Through all of it the pool is running in a degraded state, with no parity protection until the rebuild completes, which is precisely when you become very aware that you have not tested your backups recently. (I tested my backups.)
Four drives, four resilvers, the better part of four days of nervously checking progress. The fourth one finished, the new capacity finally appeared, and the deleted-files panic of last month became a distant memory. Slow, tedious, and exactly the kind of boring that storage migrations should be.