Auto-dismiss and manual close map to different importance
Aliases: close policy · auto vs manual · notice lifetime
What it is
When a notice leaves the screen is a dismiss policy, not an animation detail. Auto-dismiss hands the ending to a timer, and fits low-importance information whose absence would not change the next step. Manual close hands the ending to the person, and fits information they must admit they saw, or a situation that is still true. Policy has to track importance: the same message auto-closing today and requiring a close tomorrow destroys the model of “what a notice that looks like this means.” This leaf is how two kinds of ending map onto two kinds of information, not how many seconds one component should stay.
Why it happens
Who ends the notice is part of its meaning. A timer ending means the system considers the fact delivered; a hand ending means delivery is acknowledged to depend on the recipient. People grade messages by how they close: what walks away on its own is a receipt; what I have to close is a situation. If that grade runs against true importance—a failed payment walking off in two seconds, a “copied” pinned until someone hits the cross—the training is applied to the wrong objects. Later, a notice that needs a close is first taken as small, and a notice that walks away is not trusted to be unimportant. A policy cut by component skin (“every overlay auto-closes”) lets the skin kidnap importance, instead of letting content decide.
Where it stops holding
The same message can matter differently to different readers (license expiry visible to an admin, invisible to a member). Policy should follow importance for this reader, not the message template. An undoable-delete receipt sits between rungs: it needs to last long enough to hit Undo, but it is still an event, not a situation, so the ending can remain a timer if Undo also lives somewhere else. When law treats “shown” as “informed,” auto-dismiss is weak as evidence; switch to manual or keep a read record. When several notices mix both policies, separate them in space as well, so one slot that sometimes walks away and sometimes does not is not read as a fault.
Applying it
- Grade each class of message on importance, and lock the ending: receipts on a timer, situations on manual close or condition-cleared.
- Do not set policy by component skin (“toasts always auto-close”); set the grade, then pick the skin.
- Audit existing notices: two columns, ending versus consequence. Where they disagree, change the policy, not the tone of the copy.
- Verify by sampling ten notices, hiding the body, and asking how serious the matter is from how it disappears. A judgment opposite the real consequence means policy and importance have come apart.