I don't think we document (but that should be confirmed) that we recommend the user use unique IDs for block-based incremental backup. They're strings and can be anything the user wants. We should state that recommendation.

The IDs are stored in the metadata and survive restart and recovery. I think the only problem, if the user reuses IDs, is that WiredTiger may return backup information that data has changed (from information stored based on the previous life of the old ID) when it has not. That would imply copying data they already have which is inefficient but would not lead to corruption. So it isn't terrible.