

You can’t really easily locate where the last version of the file is located on an append-only media without writing the index in a footer somewhere, and even then if you’re trying to pull an older version you’d still need to traverse the whole media.
That said, you use ZFS, so you can literally just zfs send
it. ZFS will already know everything that needs to be known, so it’ll be a perfect incremental. But you’d definitely need to restore the entire dataset to pull anything out of it, reapply every incremental one by one, and if just one is unreadable the whole pool is unrecoverable, but so would the tar incrementals. But it’ll be as perfect and efficient as possible, as ZFS knows the exact change set it needs to bundle up. It’s unidirectional, so that’s why you can just zfs send
into a file and burn it to a CD.
Since ZFS can easily tell you the difference between two snapshots, it also wouldn’t be too hard to make a Python script that writes the full new version of changed files and catalogs what file and what version is on which disc, for a more random access pattern.
But really for Blurays I think I’d just do it the old fashioned way and classify it to fit on a disc and label it with what’s on it, and if I update it make a v2 of it on the next disc.
The full DeepSeek model is available for download, and should generate about the same quality answers as the official one, with the bonus of less censorship. I pretty trivially got it to talk about the Tiananmen Square, and they can’t even ban me for it.
That said, that’s rarely the point. It’s usually because you can, a cost saving measure, sometimes you plainly just don’t need a good model, sometimes you want privacy, sometimes you need privacy at the cost of quality.
If your business is shoving customer reviews into a model, you really don’t need the best model for it to tell you how angry the customer is.
Personally I just do it for fun and because I can. Sometimes you just do things for no other reason than because you can.