Your Test Isn't Broken. Your Pixel Is Contaminated.
A single shared touchpoint, an email click landing on the same page, a shared checkout, quietly lets exposure leak between your control and your variant. The test still runs. The result just stops meaning what it claims.
What clean actually requires
A genuinely clean test needs two groups that never touch each other anywhere in the funnel, not just in the ad account running the split. If a retargeting email goes to both groups, or the two paths land on the same checkout page, or a shared support widget nudges someone from one group toward the other, exposure has already crossed over. The ad platform only sees its own slice of the journey, so from where it's sitting, nothing looks wrong.
The leak is usually invisible
Cross-pollination doesn't throw an error or flag a warning. The test runs to completion, delivers a result, and reports a lift or a flat line with total confidence. What actually happened is the two groups had less real separation than the test assumed, and there's no dashboard indicator anywhere that says these groups touched. The number comes back clean. It just isn't measuring what the test was designed to measure.
Why an owned record catches what the platform can't
A platform-level testing tool has no visibility past its own channel. It can't see the email click, the shared checkout, or the support conversation that broke the split, because none of that lives inside its own reporting. A record built from every source, actual order history, actual visit paths across channels, can trace whether a control customer was ever exposed to the variant. That's the only way to know if a test result means what it says, instead of trusting a platform that was never watching the whole picture.