Experiments have problems too. Let's say you successfuly blind people in a nutrition study (is this possible?). Randomized. How do you ensure they're following protocol? For how long? And who is adhering? How do you know you're manipulating what you think?

Experiments introduce control at the expense of generalizability, and some types of experiments more so than others.

Observational studies aren't perfect but they do have advantages, and in nutrition I wouldnt trust experiments more (maybe not less either to be fair).

We have plenty of experiments in animals for example and those arent much more interpretable, at least in nutrition.

Nothing is simple.

> Let's say you successfuly blind people in a nutrition study (is this possible?)

What blinding do you need for all-cause mortality? Do you think people are lying on the form where the coroner says they are dead?

> How do you ensure they're following protocol? For how long? And who is adhering? How do you know you're manipulating what you think?

Every single problem here is far, far, far worse for all those observational studies you cite.

> Observational studies aren't perfect but they do have advantages

Their only advantages are that they make it cheap and easy to get the wrong answer precisely and churn out lots of papers and meta-analyses. Fortunately, here we have the more expensive, hard, but correct results. Guess what the second set of results say?

> We have plenty of experiments in animals for example and those arent much more interpretable, at least in nutrition.

Indeed. Because most nutrition research, especially the ones based on correlational results, are wrong. That's why the randomized experiments aren't 'interpretable': because the theories are wrong. You will indeed have trouble "interpreting" the results which keep showing all the theories are wrong. This is a feature, not a bug.