Specifically:

We can formalize the specs we recognize, we can signal users to be trusted to not run malicious source code. Using human trust, you can at least say that code from their events is less likely to be malicious. Computational trust, we'd need a hardened method for that.