True but how do you prove it and how do you claim an honest full download?

Because one thing is clear: the data they have on me in my archive is by far not enough to train the feed in the way it is shown to me. How do I know? Because my download archive doesn’t have that much info on me and there must be more.

I mentioned search history because it's one example where we can prove the archive doesn't export all your information. Because it's visible in the UI if you know where to look.

Presumably the same is true for likes on posts/shares, a major feed training signal. (Find the old post, it will still show that you liked it, but post likes are not in your archive.)

But yeah, then there's the darknet of data they've gathered on you that isn't visible in any UI.

I have to say: I feel pretty helpless that I can’t force them to give me that data although the EU-laws require them to give it to me (as a German).

My search history and Likes are clearly data that can be associated with me as a person and thus, fall under former EU data protection laws and for sure under the GDPR.

Couldn't you take this evidence to the bfdi[1]?

[1] https://www.bfdi.bund.de/

GPDR (assuming that's what you're talking about) doesn't take effect until next month

That is not correct. GDP enforcement doesn't start until late May, but the law is already in effect.