The observed difference in performance between variants.

Or more simply stated, the conversion rate difference between versions.

In testing, your goal is to ensure you've accurately detected a meaningful effect, or conversion difference.

To do so, you need an adequate sample size so you test has enough power to accurately pick up a real effect so you're not making an or getting results due to random chance.

For a test result to be trustworthy, the treatment effect must be larger than the Minimum Detectable Effect (MDE).