SYS/COCKPIT
CRO tools
Pre-test, post-test reading and sequential peeking — the experimentation lab in one workspace.
Lift and confidence interval
Relative lift and its 95% confidence interval, shown together. Enough to stop reading a p-value while ignoring the effect size and its range of uncertainty.
SYS/IN
SYS/OUT
Relative lift (B − A) / A
+12%
- Absolute gap (B − A)
- +0.6%
- 95% Wald CI on B − A
- -0.02% … +1.22%
- The same CI, read as relative lift
- -0.42% … +24.42%
The interval covers 0. A point lift whose CI crosses zero is not evidence that B beats A.
Wald interval (p̂ ± z √(p̂(1 − p̂)/n) on each arm, unpooled se). Not Agresti-Coull: the approximation degrades near 0% or 100%.
p-value and z: significance tool
| A | B | |
|---|---|---|
| Rate | 5% | 5.6% |
The table is the readable source; the chart is an aid.
How this is calculated
Method
Relative lift = (pB − pA) / pA.
The interval is a Wald interval at 1 − α on the difference of two proportions, unpooled se:
se = √(pA(1 − pA)/nA + pB(1 − pB)/nB), CI = (pB − pA) ± z_{α/2} se, clipped to [−1, 1].
The relative reading divides those bounds by pA. This is not Agresti-Coull.
Sources
Classical Wald on a difference of proportions. Numbers come from waldCiDifference and twoProportionZ in lib/stats. The p-value lives on A/B significance — this page shows the effect and its range.
Limits
Wald degrades near 0% and 100%. A point lift whose CI covers 0 is not evidence. The test assumes independent visitors and a fixed horizon.