FC methods don't necessarily recover all or any theoretical properties, yes. But evaluating a method against data with known theoretical properties will inform us about how to interpret the computed statistical dependencies.

I meant fMRI, yes

Matching experimental data statistics is important. But you could match the statistics without defining theoretical properties of connectivity (e.g., direction, weight). Do you think this would be useful?

what ground truth properties of connectivity should be simulated?

Well, as I live in Egypt, it's difficult to design the simulation with no exposure to experts or advisors on that topic. That's why I'm looking for a design that I could learn from or use directly.

(To clarify) My goal is to evaluate a new method for estimating time-varying connectivity from fMRI data. And to my surprise, the studies I found simulate data relationships (e.g., covariance) not underlying coupling (e.g., DCM parameters).