Experimentation
One-click winner deployment
Roll the winning variant out to everyone, keep the change live, and decide later whether to hard-code it.
When an experiment produces a winner, deploying it serves that variant to everyone instead of to a share. The change keeps working exactly as it did during the test, because it is the same change, applied by the same runtime.
A deployed variant stops being a test arm. It receives no randomised traffic, so it is excluded from significance calculations rather than being compared against a control it no longer competes with, and the experiment is treated as concluded by weight rebalancing and by automation.
Deployment is recorded: who did it and when. That record is what the implementation queue reads later to point out that a change is still being served by Pertento rather than living in your codebase.
Nothing about the rollout is one-way. The variant is still a variant, and the original is still there.