It's not inside out - a ZFS filesystem lives within each ZFS dataset (including the root one), and the root ZFS dataset can be comprised of other ZFS datasets, each with their own ZFS filesystem and properties (yet have access to all the space available in the root ZFS dataset).

For example, if you create a 'data' ZFS dataset that uses some storage device (which will be mounted to /data and contain a ZFS filesystem), then 'data' is the root ZFS dataset for that associated storage.

And if you then create a 'data/stuff' ZFS dataset, then the /data/stuff directory is a separate ZFS filesystem that shares the same underlying storage as the 'data' ZFS dataset.

This is also why the root ZFS dataset (e.g., 'data') is always displayed separately in the output of the 'zfs list' command (which displays all datasets on the system).

An APFS container is like the root ZFS dataset (e.g., 'data'), and the APFS volumes within this APFS container are like the ZFS datasets within the root ZFS dataset (e.g., 'data/stuff'). However, unlike ZFS (which can create a single dataset from virtually any number of devices/partitions/files), APFS containers are normally just created from a single big partition (slice) on the Apple SSD. And because APFS containers aren't just partitions, you can't just delete the container using 'diskutil erasevolume' - you must instead use 'diskutil apfs deleteContainer'.

Ah, I see what you mean now and this makes sense! I was overlooking the root dataset and probably slightly misusing the term 'filesystem' in a way that your usage made clear. What I was calling a 'filesystem' I should have called 'the root dataset'. So both APFS containers and APFS volumes are like ZFS datasets.

Thanks very much for the detailed reply!