Variants

A variant is one version of the page being tested. Every experiment has at least two variants: a control (the original) and one or more challengers.

A/B vs multivariate

A/B test

The simplest form — two variants, one change at a time.

50% → Control    (/pricing)
50% → Variant A  (/pricing-v2)

Best for testing big changes where you want a clear winner.

A/B/n test (multi-variant)

Three or more variants tested simultaneously.

34% → Control     (/pricing)
33% → Variant A   (/pricing-v2)
33% → Variant B   (/pricing-v3)

Useful when you have multiple ideas and want to test them in one go. Requires more traffic to reach significance.

Multivariate test (Growth plan)

Tests combinations of changes across multiple elements simultaneously. Requires more sophisticated setup and significantly more traffic.

Setting up variants

Each variant has three fields:

Name

Internal label — e.g. "Control", "Blue button", "New headline". Used in analytics events.

Target URL

The URL path your server should serve for this variant. For the control, this is the same as the base URL. For challengers, point to the alternate version:

Experiment base URL:  /landing
Control target:       /landing       (no rewrite)
Variant A target:     /landing-new   (rewritten)

The target page must already exist on your site. If the target URL returns a 404, visitors assigned to that variant will see a 404.

Traffic weight

A relative number determining what percentage of new visitors see this variant. Koryla normalizes weights automatically:

Setting a variant's weight to 0 effectively disables it without deleting it — useful if you want to pause a specific variant without stopping the whole experiment.

The control variant

The control is special:

• It represents the current version of your page (the baseline)
• Its metrics are what you compare challengers against
• If the experiment has no clear winner, you keep the control
• The control's target URL should match the experiment's base URL (so requests aren't rewritten)

URL rewriting in practice

When a visitor is assigned to Variant A, the Worker rewrites the URL server-side before the response is sent. From the visitor's perspective:

• They requested /pricing
• The URL in their browser still shows /pricing
• But the page they received came from /pricing-v2 on your server

This means:

• ✓ No redirect, no extra network round-trip
• ✓ No URL change visible to the user
• ✓ SEO-safe — only one canonical URL
• ✓ Works with any server-side framework

Sticky assignments

Once assigned to a variant, a visitor always sees that variant. The assignment is stored in a cookie:

sp_{experimentId} = {variantId}

The cookie is set for 30 days at Path=/. If you clear cookies or use incognito mode, a new variant is randomly assigned.

Is the control variant

Each variant has an is_control flag. This is used to distinguish the control in analytics events and reporting. Set it on the variant that represents your original page.

Minimum traffic requirements

As a rough guide:

For low-traffic sites, stick to A/B tests and be patient. Running a test for less than a week or stopping it early when one variant is ahead is the #1 cause of misleading results.