I hadn’t considered it - tmdjs has some of the same interfaces as TMD- what may work would be just sticking to a subset of TMD like you said.

In both cases the dataset works like a map, nth works fine on columns, etc. Building a dataset on pure clojure persistent vectors doesn't seem like a good use of time - the tmdjs system uses the js buffers such so it has js-specific bindings which are wrapped using a tech.v3.datatype interface similarly to how the jvm system wraps jvm arrays. So both systems have bindings to basically OS-specific and type-specific storage.

That makes sense.

Just the subset would be extremely valuable. Might be able to have people from our team pitch in to make it happen if you think it’s viable.

It’s definitely viable