H2Glow / Hero experiment
Drop in the impressions and conversions from your GA4 events for each hero variant. The first row is your control. Everything is measured against it.
Pull impressions and conversions straight from the Cloudflare Worker instead of typing them.
Confidence comes from a two-proportion z-test comparing each variant to the control. A variant is called a winner only when it clears 95% confidence and its conversion rate is higher. Anything below that is noise so far.
Do not stop early. Peeking and calling a winner the moment a number looks good inflates false positives. Pick a stopping point up front, for example when each variant has passed roughly 100 conversions or the test has run two full weeks, and read the result then.
Low traffic warning. Pre-launch traffic is thin, so a true difference can take a while to show up as significant. If confidence stays low with small samples, that means "keep running," not "no difference."