Stop an experiment when the result is accepted, rejected, or no longer valid. Stopping works from running or paused status. It sets the experiment to stopped, records the end time, and invalidates active experiment caches so the SDK no longer assigns new users into that experiment.

If the treatment wins

Turn the winning payload into the normal production behavior.
  • For server-driven behavior, create or update Remote Config.
  • For app-code behavior, ship the client change.
  • For purchase access, keep entitlement rules authoritative.

If the control wins

Keep the existing behavior and archive the treatment payloads. Document what you learned so the same hypothesis is not relaunched accidentally.

Cleanup checklist

  • remove any QA-only app version, country, or entitlement targeting used for launch validation
  • archive stale Remote Config payloads
  • make sure users are not targeted by two overlapping experiments for the same surface
  • keep the experiment result available for future product decisions