For AI agents: a documentation index is available at /llms.txt. Append .md to any page URL for markdown, or send Accept: text/markdown.
Ending an Experiment
Learn how to properly end experiments and make informed decisions based on your results.
Ending a Statsig experiment means choosing what happens to new users, to users already exposed, and to the experiment's results. You can ship a group to all users (Make a Decision), stop enrolling new users while exposed users keep their assignment (Stop Assignments), stop enrollment and data collection without shipping a group (Conclude Experiment and Defer Decision), abandon the experiment, reset it, or disable a single group. The following table shows which action fits each situation and where to find the details.
Choose how to end an experiment
| Action | New users | Existing users | Results | Page |
|---|---|---|---|---|
| Make a Decision | Receive the shipped group's parameter values as the new defaults | Receive the shipped group's parameter values | Remain accessible but stop updating on the day you make the decision | Make a Decision |
| Stop Assignments | No new users enter; allocation drops to 0% and new users receive the project's default experience | Keep receiving their control or treatment experience (requires Persistent Assignment) | Analysis continues for already-exposed users | Stop Assignments |
| Conclude Experiment and Defer Decision | No new users enter | Receive the default experience | Data collection stops and results conclude; you can still ship, reset, or abandon later | Conclude Experiment & Defer Decision |
| Abandon | Receive the default experience | Receive the default experience | Experiment moves to a finished state; you can reset it later | Abandon an Experiment |
| Reset | Receive the default experience until you restart; the new salt reassigns groups on restart | Receive the default experience until you restart | Experiment returns to an unstarted state and the previous run's results stay in history | Reset an Experiment |
| Disable a group | Statsig can still assign new users to the disabled group; they receive the default experience | Users in the disabled group receive the default experience; other groups keep running | Statsig continues calculating results for the disabled group | Disable a Group |
Stopping an experiment
When an experiment has a problem rather than a result, pick the action by how much of the test you still want.
- You no longer want to run this test: abandon the experiment.
- You plan to fix the problem and run the same test again: reset the experiment.
- Only one variant is bad and the other groups are still worth running: disable that group.
Reset an experiment
Resetting returns the experiment to an unstarted state, generates a new salt, and starts a new analysis window. Go to Reset an Experiment for what the reset changes and why the new run is a separate experiment.
Archive an experiment
After you make a decision on an experiment, or when you abandon or reset it, you can archive the experiment. Archiving preserves the experiment's history and results and makes the experiment read-only.
If you later decide to make the experiment active again, unarchiving the experiment makes it editable but doesn't automatically restart it.
If the unarchived experiment was previously part of a layer and you intend to include it again, you must manually reset allocation from that layer.
Was this helpful?