Hi @Fil .
After @bgchen revealed the magic of the tarball to me on this forum, I have found at least one practical use case for tarballs: Controlling View Permissions and Excerpting Sections of Long, Unwieldy Notebooks
I have one central notebook that describes a ‘case’ received by my office by way of over 500 data fields, defined as Observable inputs. For me, it’s handy to have all of these fields in a single place, but–except when using a computer with relatively large amounts of RAM and a decent processing card–I find that opening and dealing with this notebook via Observable to be pretty slow (though using @jashkenas inputs is extremely convenient). There are also some sensitivities among people I work with about providing a ‘public’ (albeit obscured) URL.
By hosting the tarball on AWS S3, I can control which IP addresses have access to the file–mitigating the second of my concerns named above. As to overcoming speed issues, I find that the JS as a tarball loads more quickly than does an Observable notebook, and it allows me the same flexibility to expose only those inputs I want in a given page. Linking the Observable inputs into a custom HTML document also affords me greater control over placement of inputs and their rendered outputs (such as creating a separate ‘input’ tab and ‘view’ tab). I do this sort of work especially when I am trying to model layouts and behaviors I want to web application developers with whom I work.
Is this the sort of use case you were looking for?