Is your feature request related to a problem?
When a team ships a winning variant, the End experiment modal offers two release modes (PR #59889): Roll out to the experiment population or Roll out to all users. The modal always starts on the first option. Some teams want the opposite default, because they run every experiment on 100% of users and want the winner live for everyone. Other teams must never roll out beyond the experiment audience. Each person who ends an experiment must remember the team rule. This is easy to miss.
Describe the solution you'd like
- Add a project setting Default ship-variant release mode with the two values. The product default stays "experiment population".
- The End experiment modal and the ship-variant API use this value when no explicit value is given.
- Show the team default in the experiment setup inspector, next to the other team defaults.
- Add a "Set as team default" checkbox in the modal, in the same way the cleanup PR repository can be set as team default today.
Describe alternatives you've considered
- Keep the per-ship radio only. This is the status today, and the problem is the forgotten choice.
- Store the last used value per user. This does not give a team rule.
Additional context
Related: #51111 (the release-mode choice, partly done by PR #59889), #54804 (do not preselect a variant when ending an experiment).
Is your feature request related to a problem?
When a team ships a winning variant, the End experiment modal offers two release modes (PR #59889): Roll out to the experiment population or Roll out to all users. The modal always starts on the first option. Some teams want the opposite default, because they run every experiment on 100% of users and want the winner live for everyone. Other teams must never roll out beyond the experiment audience. Each person who ends an experiment must remember the team rule. This is easy to miss.
Describe the solution you'd like
Describe alternatives you've considered
Additional context
Related: #51111 (the release-mode choice, partly done by PR #59889), #54804 (do not preselect a variant when ending an experiment).