Launch checks

Split Test Pro checks your experiment while you build it, not in a dialog at the end. The Experiment Summary beside the wizard tracks what is still missing, and Launch Test stays disabled until every essential field is filled in. This page lists those checks, the few that run on the server when you launch, and the manual review that catches what no automated check can.

Launch checks

The summary’s status reads Essential fields incomplete until everything required is in place, then All essential fields complete. While an essential field is missing, Continue (on the General and Targeting steps) and Launch Test (on the Variants step) stay disabled, and the summary lists what each step still needs.

General

• a name
• a hypothesis
• a default goal — or, if your workspace has no goals yet, a workspace goal (create one in Settings)
• an event name, when the test is activated by a custom event

Targeting

• a saved page rule
• at least one device
• a value for every page rule
• your open changes saved

Variants

• a variant to compare against the control
• a saved variant
• a change on every variant
• a start date, when When should it run? is set to Schedule

Warnings that never block

Some items appear as warnings: missing screenshots for a changed variant, the safety scan of your variant code, whether a variant’s rendering could be verified, and your URL targeting. They are worth reading, but you can launch with them showing.

Checks when you click Launch Test

A few checks run on the server at the moment you launch. If one fails, the test stays a Draft and the message tells you why:

• An active subscription. Without one, the message offers a Go to Billing button.
• The monthly session limit. Once usage passes the plan’s monthly session limit by the 10% grace buffer, new launches are blocked. The message points you to Settings to upgrade.
• The plan’s concurrent-test cap. Some plans limit how many tests run at once: “Your plan allows N concurrent experiments. Pause or complete one, or upgrade in Settings to run more.” See Billing and Plans.
• A primary metric. The test needs a goal to judge a winner by.

The same checks run again when a scheduled test reaches its launch time. If one fails then, the test goes back to Draft — see Scheduling Experiments.

Before You Launch

The launch checks confirm your test is complete; they can’t confirm it is right. Run through this list yourself.

1. Confirm the targeting

On the Targeting step, check that your page rules match the pages you meant. A typo in a path means zero traffic.

2. See every variant render

On the Variants step, open each variant with Edit on site and look at the change on the real page, on desktop and mobile widths. Check the variant’s screenshot too. If your CSS doesn’t apply because of theme specificity, fix it now — see CSS Variants.

3. Make sure your goal fires

The primary goal drives every winner calculation. Check that it is producing data today and that it measures the outcome you care about — testing a button colour while tracking page views is a mismatch. On an HTML site, confirm your custom event fires in the browser console.

4. Sanity-check traffic

A rough heuristic: at 500 sessions per variant per day, a typical 10% relative improvement reaches 95% probability in 5–10 days. If the targeted pages get far less, expect a longer test — or broaden the targeting, make a bolder change, or pick a goal that fires more often.

5. Look for other tests on the same page

If other tests run on the same pages, decide whether they can share visitors. After launch, the Results page shows an Overlap banner when your test shares visitors with another one — see Reading the Results Dashboard and Running Multiple Experiments.

After Launch

When the launch succeeds you see a Launched toast and the experiment opens on its Results page. Visitors who match your targeting enter the test on their next page view, so sessions should start appearing within minutes on a page with normal traffic. If sessions are still at zero after an hour, the script isn’t activating — recheck your install and targeting.

Next Steps

• Check what visitors will see: Screenshots and Preview.
• Watch for problems while it runs: Anomaly Alerts.
• Get an AI read on your results once data comes in: AI Review.