Planned changes and feedback

Hold alerts during a sale or a deploy, and tell CountWatch when an alert was wrong.

In short: before a sale, a launch or maintenance, tell CountWatch (your agent can do it) and alerts hold while it runs. When an alert comes, answer Is this real? on its page in the app: real, false alarm or expected change. CountWatch learns from each answer.

Known events. Before a deployment, campaign or maintenance, POST /v1/events with kind (and expect when the default does not fit): a level change holds both directions for up to 6 hours and then accepts the new level once steady; an expected spike or drop holds that direction until until; a pause holds drops and no-data. Held periods are left out of the usual level. watch_id: "*" covers every watch.
Feedback. POST /v1/alerts/{alert_id}/feedback with real, false_alarm or expected_change. Real and false alarms teach the weekly rule choice; an expected change on an open alert makes its new level the new normal at the next check and closes the alert.

For agents: this page as Markdown.