Ignoring the very important fact that Cassette itself still requires a bit of Base work, I’d say that Capstan is a working prototype - all the fundamental mechanisms are in place (live variable storage, tape storage, interception mechanism, etc.), but things like the user-facing API and the reverse-mode primitive catalog only barely exist. Such items should be relatively easy to implement later, though, since their design is mostly complete at this point. I’d say the bulk of the (non-Base/non-Cassette) work from here will be implementing various performance optimizations, API utilities, and tests/examples/docs.
Capstan’s simple neural net example works if you can find the magical Base + Cassette commits on which things currently run.
As far as a timeline goes, the (very) tentative idea is that Capstan will have an initial release around the same time as Cassette, and Cassette will have a release once Base can support it, and Base’s support for Cassette is on the v1.1 milestone, so…stay tuned!