Stop peeking.

You sized the test, then watched it like a hawk and called it the moment it crossed 95%. Every extra glance is another chance for pure noise to wander over the line — so the false-positive rate you actually ran at is far higher than the one you chose. This is how much higher.

I call it at% confidenceThe per-look significance threshold — how sure a single test would need to be before you'd call a result real. 95% confidence means a 5% false-positive rate, if you only look once., and peek at the resultstimes.
5 looksAn interim analysis: a point during the test where you stop and check significance, ready to call it early if it has crossed the line. Equally spaced here, the way a daily dashboard check would be. at α = 5% (2-sided), z* = 1.960
14.2%True false-positive
5%You intended
2.8×Inflation
Badly inflatedWith 5 looks your real false-positive rate balloons to 14.2% — 2.8× the nominal 5%. Most "winners" you stop on early are noise.
If you peek…Real false-positive rate
once
5.2%
2 times
8.1%
3 times
10.9%
5 timesyou
14%
10 times
19.6%
20 times
24.8%

Check often enough — a live dashboard you refresh daily — and a test with no real effect is almost guaranteed to flash "significant" at some point. Monitoring continuously drives the true rate toward 100%.

The blunt, always-safe patch is BonferroniSplit your error budget evenly across the looks: test each at α ÷ (number of looks). Simple and never too lax — but conservative, since consecutive looks are highly correlated.: test each look at 99% confidence instead, so all 5 looks together still spend only your 5% budget.

Better still, use a method built for it — group-sequential boundaries (Pocock, O'Brien–Fleming) or always-valid sequential tests — which let you watch continuously without inflating the rate, rather than just spending the budget more frugally.

repeated-significance simulation · the cousin of A/B Test & how not to run an A/B test