Current behavior for GB experiments is an experiment started before 14:30 UTC waits until 14:30 UTC to start. An experiment started _after_ 14:30 UTC starts immediately, and has the time backdated to 14:30 the same day.
This is not ideal for a few reasons:
- GB's UI shows information that is in fact not what's happening in the system
- GB isn't respecting the notion of a deployment window, which is specifically intended to limit the time(s) that production behavior can change. Other teams are affected by these changes.
Acceptance Criteria:
- Update the behavior to start during the window:
- If start before 14:30 UTC, experiment should turn on at 14:30 UTC that day
- If start between 14:30 and 15:00 UTC (i.e. the end of the deployment window), it can start immediately -- it's within the window
- If started after 15:00 UTC, the experiment should turn on in the next deployment window and the start date is updated to reflect this.
- The same behavior applies to experiments which are scheduled to start as well as those started immediately with the start button.
- Update the item in the pre-launch checklist about start dates to include acknowledging that the experiment start time will be during the next valid deploy window