Experiment Reviews

Submit, approve, and commit Statsig experiment changes in a project that requires config reviews, including self-approval and how to clear a pending review.

When a project requires config reviews, Statsig holds every experiment change until someone approves and commits it. Buttons that normally apply a change read Submit for Review instead, the experiment's Setup tab becomes read-only while a change is pending, and the header hides Start or Make Decision until the change clears. This page covers what changes for an experiment. To turn reviews on for a project or an environment, go to Set up reviews.

When an experiment change needs review

A change needs review when both of the following are true:

The experiment requires review. Statsig requires it when the owning team requires it, the project requires it, or the experiment itself requires it. Any one of these is enough; a level that doesn't require review can't cancel one that does.
The change affects an environment that requires review. Environments scope where a requirement applies, but they never create one. Production always requires review; other environments don't by default.

Warehouse Native analysis-only experiments are exempt unless the project also requires reviews on analysis-only experiments.

Submit a change for review

Make the change as usual. For a lifecycle action, open Start or the Make Decision dropdown and choose the action you want.
Optionally, add reviewers in the Changes require review block. You can add several reviewers, but only one of them needs to approve. Statsig adds reviewers automatically when the change touches an environment that requires a specific review group.
Optionally, describe the change.
Select Submit for Review.

A banner then appears at the top of the experiment naming the request, such as "my_experiment was requested to be started", along with who requested it and which reviewers it went to. Statsig emails the reviewers and notifies them in the console, and in Slack if you've configured the Slack integration.

Which experiment changes are reviewable

Reviews cover the experiment's whole lifecycle, not just launches: creating, starting, and scheduling a start; making a decision, including ship, rollout, and scheduled rollout; stopping assignments, concluding, abandoning, resetting, and restarting; scheduling or editing a stop; archiving, unarchiving, and deleting; and editing settings, overrides, owners, team, target applications, or allowed reviewers.

Approve and commit a change

Approving a review doesn't change the experiment. A reviewer approves first, and then someone commits.

A reviewer opens the experiment, selects View Changes to compare the current and proposed configuration, then selects Approve or Reject.
The review moves to Approved & Awaiting Commit.
Someone with edit access selects Commit Changes to apply it.

Statsig doesn't notify anyone when a review is ready to commit. Approved changes stay pending until someone commits them, so check the project's Reviews page or the experiment banner for work in the Approved & Awaiting Commit state.

You can't approve your own review.

Commit without review

If your role has the Self-Approve Review permission, your own pending change shows Commit without Review instead of Approve. Selecting it asks you to confirm, then applies the change immediately and records it in the experiment's history as self-approved.

This permission is on by default for the Admin role and off for Member, so turning reviews on doesn't by itself produce two-person approval. To require a second person, turn off Self-Approve Review for every role that has it.

Withdraw or clear a pending review

Before approval, the requester selects Edit > Cancel Submission.
After approval but before commit, select Discard. Anyone with edit access can discard a pending change, not only the requester.
After commit, make a reversing change. You can't cancel a committed review.

An experiment holds one review at a time. Submit a second change only after the first one commits or clears.

When the experiment changes under a pending review

A review stores a snapshot of the experiment as it was at submission. If anything edits the experiment afterwards, committing fails with a message saying the experiment has been modified since your change. Refreshing doesn't clear this, because Statsig replays the stored change. Check the experiment's history for what changed, discard the pending review, make the change again, and commit it promptly.

Allowed Reviewers

The Details panel shows an experiment's Allowed Reviewers. When this list isn't empty, it limits who you can pick as a reviewer when you submit a change. Editing the list is itself a reviewed change, routed to the existing reviewers, so an editor can't remove the people who review their work.

Was this helpful?