Monitoring
Email alerts
On for the things that genuinely warrant interrupting you, off for the things you just did yourself.
Email is the channel that reaches somebody who is not looking at the console, which makes it the right place for the events that need a person and the wrong place for the events that are simply a record of what happened.
So the defaults split along that line. Health signals, a split mismatch, a dead experiment, an overdue one, a blocked automation, are on. So are results: a significant positive and a significant negative both warrant knowing about today rather than next week. Lifecycle events, an experiment starting, ending or being deployed, are off, because the person reading about them usually just did them.
Those defaults apply to any account that has never opened the settings screen, which is most of them, and they live in one place rather than as column defaults in a table. Changing the policy later therefore actually moves every account that never touched it.
Delivery never affects the action. Email is sent after the event is committed and outside the transaction that recorded it, so a slow mail provider can never roll back something that genuinely happened.