Honest statistics for small samples
Four free tools built on one idea: honest inference on small, noisy data. No p-values that break when you peek, no false precision, no dashboards that quietly declare winners — just what the data can and cannot tell you, with the uncertainty shown. Free, no signup, and they run entirely in your browser.
Is my ad result real?
Bayesian A/B for $50-a-day advertisers. Get "84% probability B beats A," a credible interval on the uplift, and a straight call: keep running, pick a winner, or no real difference yet. Safe to peek.
Is my survey actually valid?
Paste your questions and get them graded for leading, loaded, and double-barreled wording — with rewrites. Then turn raw responses into proportions with correct confidence intervals and honest "too small to conclude" warnings.
Is my idea validated?
Log your discovery signals. A stranger paying money counts; a friend saying "cool idea" barely does. Get a calibrated confidence score, bias flags, and the single next conversation that would move it most.
Will my drop actually sell?
Smoke-test your small-batch merch before you make a unit. Turn landing-page signals into a projected paid conversion with an interval, fold in real fabrication cost, and get a go / no-go on the production run.
Why one spine, four tools
Whether it's an ad test, a survey, a founder's interviews, or a merch drop, the underlying problem is identical: drawing an honest conclusion from a small, noisy sample. These four share a single, unit-tested statistical core — see the self-tests — so work on one sharpens all of them. They're the internal instruments behind Kerf and Code's own products, made small enough to hand to anyone.