Thank you for the reply.
Oh, sad - I was begging that the use of variables is possible...
Sure, I tried to create a quick demo project illustrating my issue:
I use projects, and tasks.
I have a CSV file for all
projects ccontaining all project attributes (about 50 attributes).
I have separate CSVs for each
task type (about 150 attributes).
So my test structure looks like this:

- 2020-08-25 09_43_16 -Window.png (13.54 KiB) Viewed 207 times
The "parent-level" data source iterates through all projects.

- 2020-08-25 09_44_01 -Window.png (24.57 KiB) Viewed 207 times
The "child-level" data sources iterate through the different task types.

- 2020-08-25 09_46_46 -Window.png (26.81 KiB) Viewed 207 times
These screenshots show the previous configuration, where I had the project ID in
all data sources and a double condition on each of the child folders, one to match the project ID in parent data source, and one to match it in child data source.
But I want to get rid of the hardcoded projectIDs, they should only be in ONE csv (data_project.csv) for maintainability, that way they have to be changed only in one place and not 4.
So the idea is, remove the ProjectID from all child data sources, and set the child-conditions to match the task type, which is currently processed by the parent data source.
Without variable, i can only hardcode this for every type, where i end up like this for 3 different project types: