How about a test-compile mode where any of the 'try’s in the source code may fail with a given probability, or a batch test run mode where each of them will fail at least once? Compile with these enabled to run a test suite where all your library code is acting funky and make sure you get the top-level response (and inner local state melange) you expect?
I think the closes right now is using the test with the fuzzing framework, it takes some times to setup, but it’s very helpful in finding those weird edge cases, and with the web interface, you can see in real time branch hit.