# Experiments

**Experiments** owns drafting, publishing, measurement, and decisions. Page inventory and Opportunities live in Pages, automated behavior lives in Autopilot, and configuration lives in Project settings.

```serpclimber-doc
{
  "type": "projects-experiment-states"
}
```

## Draft generation

Starting an Experiment creates a Draft generation job. Generation uses synchronized Page content, Page keywords, effective Project settings, and any bounded research or steering supplied for that Draft. A queued or running job is shown as **Generating** before a Draft record exists.

In **Draft**, review:

- The Experiment Page and primary keyword.
- Metadata and any slug proposal.
- The structured content diff against the baseline.
- The hypothesis, evidence, research provenance, and risks.
- Whether the Experiment will publish automatically through Wix or use a guided checklist.

Chat inline edits apply to the current unpublished Draft variant, not the captured baseline. They can target one section or the whole Draft and always prepare one atomic confirmation with a frozen current-to-proposed diff. If the Draft changes before confirmation, the edit fails without a partial write and must be prepared again.

Edits and regeneration can overwrite Draft content. Review the complete confirmation diff before applying an edit; use Regenerate when the requested whole-Draft rewrite exceeds the inline-edit safety limits.

## Publishing and running

With a connected Wix resource, publishing performs a fresh preflight and may require a slug decision. It writes the approved variant through that project binding and starts measurement. External publishing, indexing, and traffic exposure can remain even if a later action restores content, so publishing is conditionally reversible.

Without a supported content-host connection, the Experiment uses **guided** mode. SERPclimber provides a checklist and the complete content changes to apply on the site yourself. After applying them, confirm **I made these changes — start measuring** and optionally supply the real application date, up to 90 days earlier. That confirmation starts measurement; a follow-up crawl checks for the changes but remains advisory because some sites render content in ways the checker cannot see.

While **Running**, SERPclimber compares current evidence with the Experiment baseline and shows data freshness. Charts and table trends include the latest available GSC activity; a dashed tail indicates the most recent values that may still change. Rates and automated decisions continue to use settled full-day evidence. The decision window comes from the Experiment or its effective settings; it is not a universal fixed duration.

## Awaiting decision and completed

When the configured decision point is reached, an experiment may briefly show **Scoring** while GSC finishes the fixed evidence window. SERPclimber does not extend that window or use later observations. Once the window is ready, review the evidence before confirming an outcome. Depending on the application mode and selected action, SERPclimber can keep the published Draft, restore a previous Wix version, record a guided outcome, or mark the result inconclusive. The confirmation explains the exact effect available for that Experiment.

After a decision, the experiment is **Completed**. The same experiment ID and history remain available, and the page may become eligible for a future iteration after its cooldown and other rules permit it.