Myth: button color is a meaningful lever. False — across thousands of tests it's near-zero median impact.
Meta-analyses of public test repositories put median CTA-color uplift inside the noise band: roughly 0% ± your minimum detectable effect (MDE).
Contrast against the page matters; the specific hue does not. Spend the test slot on copy, friction, or offer instead.
Color ▮ vs friction-removal ▮▮▮▮▮▮
Read the number, not the story. [n>1k tests · ~0% median]