Today, this (normally/usually?) only fires against reviewers that the user actually typed in using their real keyboard, so I think we have some kind of reasonable-ish expectation that they believe those reviewers are currently (or soon) active.

It also fires before we actually create a revision, so Herald hasn't acted yet and hasn't added packages/projects/etc. We could fire this later in the flow, but then if the user says "n" ("no, let me pick different reviewers") we're in some amount of trouble since we already created the revision.

If this was regularly firing against a bunch of random auto-filled junk (today or in the future) I generally agree that we'd want to move toward making a stronger effort to raise this only if it's actionable. And it's possible that we'll want to move in that direction for whatever reason.

We could also do stuff like ignore busy states ending in the next 1-2 hours (e.g., prompt probably isn't useful if it's just telling you that I'm currently in a meeting for the next 20 minutes), and show the user what events the reviewers are attending (if they can see them).

T731 might motivate some of this, e.g. a "At least 3 reviewers must approve this change" with only 1 reviewer. But maybe just dumping that kind of stuff as warnings at the end of the output would be good enough? I guess we can cross that bridge when we come to it.