What should I run if I want joblog to be in JSON/CSV but output to be in files?

Maybe extend --joblog to include column Stderr+Stdout if output is to a file?

parallel --jl foo.csv --files ...

What happens if you combine?

parallel --jl foo.csv --results bar.csv ...

This way --joblog foo.csv and --results foo.csv will do the same except --joblog will never have the actual output (at most a reference to a file), whereas --results will have the full output.

--files must still list the files on stdout so this is possible:

    parallel --files sort ::: file{1..100} |
      parallel -Xj1 'sort -m {}; rm {}'