Auto-pause

A rate alone is not enough to act on. Pausing on a rate computed from a handful of events takes a working mailbox out of service because of noise — one complaint can be a mis-click on a campaign nobody else objected to.

So a sound trigger requires both: the rate over the limit, and enough absolute events for the rate to be meaningful. Two complaints is a pattern; one is an incident.

A pause that clears itself re-runs exactly the conditions that caused it. The rate that triggered it is computed over a window, so time alone will eventually drop the offending events out of the sample and the mailbox will resume sending into the same list with the same copy.

Requiring a person to resume forces the only useful question — what changed? — to be asked. It also creates a record of who decided the risk was acceptable, which matters when the answer turns out to be wrong.

Look at the cohort that produced the events, not at the mailbox. A bounce spike almost always traces to one imported list; a complaint spike almost always traces to one sequence or one audience.

Then suppress what needs suppressing, fix the source, and resume with reduced volume so the next window's evidence is built slowly. Resuming at full cap into an unchanged list simply schedules the next pause.

SalesShift pauses a mailbox at a 0.3% complaint rate with at least two complaints, or a 5% bounce rate with at least five bounces. An auto-pause never clears itself — resuming is an explicit human action.