Think of jq more like bash instead of a generic functional language. Blocks work on data and you pipe data along blocks. You can name blocks. That 99% of what you need to get to do jq.
As you can see it loads data from vars, fed by other jq scripts. Don't try to put everything in one script. Each script should basically just reduce and normalize.
Here's a small example of a smaller ETL script a wrote a while ago for stats collection over scm repos: https://gist.github.com/turbo/36e87947a56cfaacec9d0356b3e521...
As you can see it loads data from vars, fed by other jq scripts. Don't try to put everything in one script. Each script should basically just reduce and normalize.