Sort of an aside for the general decision making for this process: for any effort put into this, I think it should be considered against the opportunity of using that effort to write better unit/integration/example tests for the interface in question. Ultimately I think better tests are always going to do the best job of backstopping breaking interface changes. At the very least they make it a conscious choice rather than an accident.