> For the complete documentation index, see [llms.txt](https://docs.elevateab.com/elevate-helpcenter/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.elevateab.com/elevate-helpcenter/experiment-setup/experiment-lifecycle.md). # Experiment Lifecycle *** Every experiment in Elevate A/B Testing moves through five states: Draft, Preview, Running, Paused, and Done. The state tells you whether traffic is being split, whether data is being collected, and which actions you can take. ### State summary | State | Traffic split | Data collected | What visitors see | | ----------- | ------------- | -------------- | ------------------------------------- | | **Draft** | No | No | The original experience | | **Preview** | No | No | The original experience | | **Running** | Yes | Yes | Control or a variation, by allocation | | **Paused** | No | No | The original experience | | **Done** | No | No | The original experience | Only Running affects your storefront. In every other state your store behaves exactly as it did before you created the experiment. ### How states connect ``` Draft → Preview → Running → Done ⇅ Paused ``` Paused returns to Running. Done can be resumed back to Running. ### Draft The experiment is saved but not finished. Nothing is live. Use Draft to configure an experiment across several sittings, or to hold a plan you are not ready to run. Drafts have no effect on your store and no expiry. **Available actions:** Edit, move to Preview, Delete. ### Preview The experiment is fully configured and ready to be checked before it goes live. This is where you verify that each variation renders correctly, on the devices and templates your visitors actually use. No traffic is split in Preview. You are the only person seeing the variations, through the Preview mode for your experiment type. **Available actions:** Launch (moves to Running), Mark as Complete, Duplicate, Delete. Preview and QA steps differ by experiment type. See the Preview and QA page inside your experiment type's section for the checks that apply. ### Running The experiment is live. Visitors are being assigned to control or a variation, and results are updating as data arrives. While an experiment is Running: * Visitors are split according to your traffic allocation, after audience targeting and traffic isolation have been applied * The Web Pixel records views, add to carts, checkouts, and purchases against the assigned variation * A visitor stays in the variation they were first assigned on every return visit * Results and analytics update continuously **Available actions:** Pause, Mark as Complete, Duplicate, Delete. #### Editing a Running experiment Changing an experiment's configuration while it is Running changes what later visitors see without changing what earlier visitors already saw, which mixes two different experiences into one result set. Where a change is unavoidable, treat the experiment as restarted and read the results from that point forward. ### Paused The experiment is temporarily stopped. New visitors are not entering it, and everyone sees the original experience. Nothing collected so far is lost. **Pause when:** * You have found a problem with a variation and need time to investigate * A sale, promotion, or campaign is about to distort buying behaviour * You are making store changes that would affect the pages under experiment **Available actions:** Resume (moves back to Running), Mark as Complete, Duplicate, Delete. #### Does pausing damage my data? No. Paused time is excluded from the calculations that depend on elapsed time: | Calculation | How pausing affects it | | ------------------------ | --------------------------------------------------- | | Experiment duration | Paused days are not counted | | Projected revenue | Visitor velocity is measured over running time only | | Statistical significance | Only data from running periods is included | Pausing costs you calendar time, not data quality. The exception is a long pause. Over several weeks, seasonality, traffic mix, and pricing across your store all shift, so the visitors before and after the pause are not comparable in the way two arms of the same experiment need to be. For pauses beyond a couple of weeks, end the experiment and start a fresh one. ### Done The experiment has finished. Traffic splitting has stopped and all visitors see the original experience. Results, reports, and raw data remain fully available. Reaching Done does not apply anything to your store. If a variation won and you want to keep it, that is a separate step. See [Ending an Experiment](https://claude.ai/elevate-helpcenter/operate/experiment-setup/ending-an-experiment). **Available actions:** Apply Winner (where the experiment type supports it), Resume, Duplicate, Delete. #### Resuming a completed experiment Resuming returns the experiment to Running and continues from where it stopped. Existing variation assignments are preserved, so returning visitors stay in the arm they were in. Resume when you ended too early, or when results were inconclusive and more data would settle them. Do not resume to give a losing variation another chance after seeing the results: choosing to keep collecting because you dislike the current answer is how a null result eventually turns into a false positive. ### Deleting an experiment Deleting is permanent and is available from any state. When you delete an experiment: * It is removed from your dashboard * Its collected data is removed and cannot be recovered * Its configuration is removed from your store's metafields * Any variation content created inside Elevate A/B Testing for that experiment is removed > **Warning:** Deletion cannot be undone. If you want the experiment off your dashboard but want to keep the findings, mark it as Complete instead. Completed experiments keep their full results indefinitely. ### Where the state is shown | Location | What you see | | --------------------------------------- | ----------------------------------- | | Experiment card on the dashboard | Colour coded state badge | | Top of the experiment detail page | Current state | | Experiment History, on the Raw Data tab | Every state change with a timestamp | Use Experiment History when results look inconsistent. A pause you forgot about, or a launch later than you remember, usually explains a duration or velocity figure that looks wrong. ### Next steps * [Ending an Experiment](https://claude.ai/elevate-helpcenter/operate/experiment-setup/ending-an-experiment) for what to do once an experiment reaches Done, and how to keep a winner * [Audience Targeting](https://claude.ai/elevate-helpcenter/operate/experiment-setup/audience-targeting) for which visitors become eligible before a variation is assigned * [Traffic Isolation](https://claude.ai/elevate-helpcenter/operate/experiment-setup/traffic-isolation) for limiting what share of eligible visitors enter a Running experiment * [Statistical Significance](https://claude.ai/elevate-helpcenter/analyze/reports/statistical-significance) for judging whether a Running experiment has an answer yet