I think when doing integration testing with ToxiProxy often people want it to be "predictable". Most of the time you just want to prove the application behaves correctly when it needs to reconnect. Otherwise you can get tests which sometimes fail and sometimes not. That's also a valid test, but then it's a smoke test.

What I often do is create a small framework which uses a pseudo random number generator for all the "randomness". I then feed a list of seeds to this random number generator which makes the tests "random" but repeatable since you know the seed.