Also given that there are huge differences in complexity between tests, how do we know that the successful 22% are not just trivial one-liner tests?

Thinking about the test suite in my current project there is a clear Pareto distribution with majority of tests being simple or almost trivial.