Installation quickstart

WordPress: use Experiment Connector for no-code heading, paragraph and button-text tests. JavaScript: use the browser snippet for static or dynamically rendered website text. PHP, Laravel and Symfony: render the JavaScript installation in your shared head template. React and custom apps: use the packaged SDK to render variations in your own code; the React adapter has limitations documented in its guide.

Content experiments replace safe, uniquely matched text. Developer-controlled experiments provide an assignment and tracking, not remotely supplied application code. All integrations use the same accounts, experiments and results.

Create your account, open Websites, add your HTTPS domain and open its installation page. WordPress proves ownership through pairing. Other integrations use a DNS TXT record or an HTTPS verification file. Copy the public site identifier from your own website page; example placeholders in these guides must be replaced.

The public site ID is not a credential and cannot verify domain ownership. You never need a private database key or installation secret in browser code.

The current browser bundle and WordPress source are version 0.5.0. The React source package manifest is version 0.4.0. Both developer packages now build installable npm-format archives with JavaScript and TypeScript declarations, but neither is published to npm. Install local archives from authorised source access until a registry release is approved.

Checked on 17 October 2026: the published /sdk/0.5.0/testwell.js address returned HTTP 404. The asset exists in this project preview but this release needs publication before customers can install it from the published URL. Do not deploy an integration pointing at an unavailable release.

In Experiments, choose website content or developer-controlled. Set original and alternative, a single primary goal and launch only on your verified website. The split is 50/50. Developer tests also need the exact experiment key in your application code.

On a staging site, grant the applicable consent, check the configuration request, confirm the presented variation and its exposure, then complete the real goal and check recorded results. An empty experiment list is normal before launch. A successful script download alone does not prove an experiment works.

Repeat page loads to check assignment persistence. Exercise navigation, failure and consent withdrawal. Disable website testing and confirm restoration. Never label unexecuted production checks as passed.