You work on an online product team and an A/B test starts showing results that look directionally promising, but the traffic split and some event patterns do not match what you expected. You suspect the experiment may have integrity issues rather than a true product effect.
How would you handle sample ratio mismatch or other experiment integrity issues in this test, and how would you decide whether the results are still usable?
