Understand conversion lift, sampling uncertainty and next actions from one transparent workspace.
Explore experiments with explainable statistics, realistic confidence bounds and a reproducible planning lab. Built for thoughtful decisions, not significance chasing.
Observed group rates with Wilson score intervals
Select any experiment to update the dashboard
Quickly compare your sample studies, organized by observed lift
| Experiment | Visitors | Lift | p-value | Assessment |
|---|
Compare conversion rates, explore uncertainty and estimate the sample size needed for a future experiment.
Controls apply to the selected analysis
Per-variant sample size target
Independent two-proportion z-test for the selected sample
Hypothetical treatment outcomes, keeping group sizes fixed
Scenario results are hypothetical. Moving the slider does not alter saved measurements or establish a real treatment effect.
Create, edit, archive and restore experiments. Everything stays in your browser.
All calculations are performed locally. Review the estimands and assumptions before relying on an analysis.
Conversion rate is successes divided by visitors. Absolute lift is treatment conversion rate minus control conversion rate (percentage points). Relative lift is absolute lift divided by control rate, when nonzero. Relative lift can be undefined if control conversion is zero.
The fixed-horizon two-proportion z-test compares a pooled conversion rate under the null of equal probabilities. The p-value is a measure of data compatibility with the null, not the probability that the treatment works. Small expected cell counts invalidate this approximation.
Each group uses a Wilson score interval for binomial proportions. The difference in proportions uses the Newcombe hybrid score interval, based on Wilson endpoints. These intervals describe uncertainty under independent Bernoulli sampling; they are not causal guarantees.
Balanced sample size follows the normal-approximation two-proportion formula with z-scores for the selected two-sided significance level and target power. This is an approximate pre-test design aid, not post-hoc achieved power.
Do not repeatedly peek at a fixed-horizon p-value, stop opportunistically, or change outcomes mid-test. Check randomization, assignment integrity, measurement quality, guardrail metrics, multiple comparisons and business impact. Low p-values alone are not grounds to ship.
Only synthetic data are included. LocalStorage and downloaded JSON/CSV are stored on your device, not synced to a server. Browser storage is not encrypted. Avoid confidential production information. No external dependency or API is required.