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

# Change the baseline

> By default the Original variant is the reference. Override it when you want to compare two treatments head-to-head.

Every result is a comparison to a baseline. The baseline is one
variant, treated as the reference, lift, CI, and verdict are all
relative to it.

By default the Original variant (position 0) is the baseline. That is
usually right. Occasionally it is not.

## When to change it

* **You are running Original, A, B, and A is your current champion.**
  Comparing B against Original mixes two changes. Set A as the
  baseline; the page now compares B to A directly.
* **The Original is a placeholder.** Sometimes the Original is
  synthetic, a control built specifically for the test that no real
  visitor sees. Set one of the treated variants as baseline.

## How

On the results page, click the **Baseline** dropdown in the meta bar
at the top. Pick a variant. Everything on the page recomputes with
that variant as the reference.

The URL updates to include `?baseline=<variantKey>`. Share the URL
and your colleague sees the same view.

## Resetting

Next to the dropdown, a small **Reset** chip appears once you have
overridden the baseline. Click it to return to Original.

## What changes

* **Lift numbers.** Every variant's lift is now vs the new baseline.
* **Verdict copy.** "Variant B wins" becomes "Variant B wins vs
  Variant A".
* **CIs.** The confidence intervals are computed as differences from
  the new baseline.
* **Charts.** The over-time confidence chart plots the new baseline
  as zero.

## What does not change

* **The data.** Every event is still counted the same way.
* **The test itself.** No visitors are re-bucketed; the baseline is a
  display choice only.

## With segments

Baseline override composes with [segments](/results/segments), pick
a segment first, then change the baseline, or vice versa. Both end
up in the URL and share cleanly.
