This is so promising!

I presume like pandoc, the goal is to use an internal model to reduce the M*N complexity of format transcoding to M+N.

But when I look at the tests for this (and Dasel and rq also mentioned here), I see very little. jq seems to have much more. Perhaps I'm missing them.

Which leads to some side/related comments...

A great project would be to build a transcoded test suite that all these tools could run against. It wouldn't be hard to generate.

Also, while go is a great language, I think pandoc gets a lot of benefit from implementing the internal model in haskell. Pandoc has a much harder problem because there's a lot more variability and less structure in the formats, so it should be doable for formats.