A/B Testing

A/B testing in VersionR lets you split SDK traffic between two prompt versions and track which one performs better — without changing your application code.

How it works

1. You configure a split on a prompt: e.g. 80% version A, 20% version B
2. The SDK fetches a version transparently — each call gets routed to A or B based on the split
3. Results logged via pd.log() are tagged with the variant
4. The dashboard shows performance metrics per variant side-by-side

Your application code doesn’t change. The pd.get() and pd.render() calls look identical — the routing is handled server-side.

Setting up a split

From the prompt detail page:

1. Click A/B Test in the version toolbar
2. Select the two versions to compare
3. Set the traffic split percentage (e.g. 80/20)
4. Click Start test

The test runs until you stop it or declare a winner.

Metrics

You can measure:

The score field gives you flexibility — populate it from user ratings, downstream model evaluation, or any signal that matters for your use case.

Declaring a winner

When you’re satisfied with the result:

1. Open the A/B test view in the dashboard
2. Click Promote to 100% on the winning variant
3. The losing variant is archived

Traffic immediately shifts to 100% of the winning version. No code change required.

Design principles

• The SDK is unaware of A/B testing — routing is entirely server-side
• No SDK update is needed when starting or stopping a test
• Both variants use the same slug — your code never needs a branch
• A test can be paused and resumed without losing accumulated data