Reset an Experiment

Reset a Statsig experiment to return it to the unstarted state, generate a new salt, and run a fresh iteration without carrying over the previous run's results.

Resetting an experiment returns it to the unstarted state so you can fix a problem and run the test again. Statsig generates a new salt, creates new group IDs, clears the start and decision state, and starts a new analysis window. Reset an experiment when a bug or a bad variant invalidated the run and you want the same test to start over.

Use a different action when you don't need a fresh run:

To end the experiment without running it again, abandon it.
To turn off one bad variant while the rest of the experiment keeps running, disable that group.
To begin a new analysis window without re-randomizing anyone, restart results. A restart keeps the existing salt, so users keep their group assignments.

Reset an experiment

Open the experiment and select the Make Decision dropdown.
Select Reset.
Start the experiment again after you fix the issue that prompted the reset.

What happens after you reset an experiment

The experiment returns to the unstarted state. Every user receives the default experience, either the value defined in code or the layer default, until you start the experiment again.
Statsig generates a new salt. Because the salt determines which group a user hashes into, users are randomly reassigned when you start the experiment again and don't necessarily land in the group they were in before. This reassignment keeps a bug or bad experience in the previous run from affecting the same group after you address the issue.
Statsig creates new group IDs and preserves each group's name and size.
Statsig clears the analysis window and saves a snapshot of the results collected so far.
You can archive the experiment after a reset to preserve its history and make it read-only.

Treat the new run as a separate experiment

A reset creates a new iteration rather than undoing the previous one, and Statsig can't restore an experiment to its pre-reset running state.

Results from the earlier run stay visible in the experiment's history, but don't combine them with the new run as a single scorecard. Mixing a pre-reset period with a post-reset period introduces assignment artifacts that bias the comparison.
Setting the salt back to its previous value doesn't resume the earlier run. It brings future assignment closer to the old hash mapping, but the analysis window has already restarted.
A reset doesn't restore allocation on an experiment whose allocation was set to zero. Raise allocation as a separate edit after you reset.

Was this helpful?