> What are the algorithms that can dismiss “indirect correlation”?

Anything that doesn't get stuck on local maxima do.

> How can you tell a direct from an indirect correlation in data?

There is another variable, with stronger correlation, that explains the same change.