> ## Documentation Index
> Fetch the complete documentation index at: https://docs.aftersell.com/llms.txt
> Use this file to discover all available pages before exploring further.

# How do tests work?

> Create and run Aftersell Cart A/B tests, read the significance indicator, and learn what can and can't be tested.

A **cart test** is an A/B test between two whole carts. You test your live cart against one challenger, split traffic between them, and pick the winner when you're done.

## What you're testing

The unit of a test is a **complete cart** — an entire experience, not a single block or setting. You compare two of your saved carts wholesale. Any draft cart can be the challenger.

## Creating a test

From the **Carts** page, choose **Create test**. You'll need a live cart and at least two carts total. Then:

1. Your **live cart is the control** automatically — you can't change it.
2. Pick **one challenger** from your other carts.
3. Set the **traffic allocation** — the split between the two, defaulting to 50/50. Each side is between 1% and 99%, and the two must add up to 100%.
4. Give the test a **name** (required).
5. **Publish & start test.**

Only **one test can run per store at a time.**

## While a test runs

* Shoppers are **stickily bucketed** into one variant, so a returning shopper keeps seeing the same cart for the length of the test.
* **Publishing and unpublishing are locked**, and the **variant carts are read-only** in the editor — end the test to make changes.

## The significance indicator

While a test runs, Aftersell shows a **significance progress bar**: the impressions collected so far against a target sample size. The target is a **total across the test**, sized so the *smallest* variant still gathers about 3,500 impressions — roughly **7,000 in total** for a 50/50 split, and proportionally more for an uneven one (an 80/20 split needs about 17,600). As the test runs, impressions accumulate and the bar fills toward that target.

The bar appears only while a test is live. Once you end the test the bar disappears; the banner keeps showing the test's start and end dates, and you read the outcome from the per-variant results.

<Note>
  The bar is a **readiness indicator** — it tells you whether you've gathered enough traffic to make a confident call. It is **not** a statistical p-value, and Aftersell does **not** auto-declare a winner. Let the test gather enough data, then compare the variants and choose.
</Note>

## Ending a test and choosing a winner

When you end a test, you see each variant's **Revenue**, **conversion rate (CCR)**, and **impressions** side by side, and you **pick the winner yourself**. The winner is published live and the other cart returns to draft.

<Note>
  **The end-test Revenue figure is broader than the one on the Test analytics page.** Here it's every conversion attributed to the cart. On [Test analytics](/aftersell/cart/test-analytics) the Added revenue metric is narrowed to upsell, add-on and subscription revenue. Expect the two numbers to differ for the same test, and compare variants within one surface rather than across them.
</Note>

## What can and can't be tested

* **Can:** any two whole cart configurations. Swap layouts, blocks, copy, design — build a challenger cart however you like and test it against the live one.
* **A caveat on metrics:** the Added revenue metric on Test analytics covers upsell, add-on, and subscription revenue — a subset of everything a cart earns rather than a full partition. Treat it as a like-for-like read between the two variants rather than as each variant's total contribution. The end-test screen uses the broader figure described above.

See [Test analytics](/aftersell/cart/test-analytics) for how to read the results.