The life cycle of an alerts channel is predictable. It is created with good intentions, it fills with everything, someone mutes it on a Friday, and the incident that mattered lands there on Sunday unread. The channel did not fail. The decision about what to send did.
Three decisions keep a channel worth reading. A minimum severity per channel, so the info-level chatter never reaches the one that pages. Quiet hours that hold warnings until morning while letting critical alerts through, so the 3am message is always one that deserved 3am. And retries with a limit, so a flaky delivery does not become a flood.
The fourth decision is the hard one: which conditions get to be alerts at all. The honest answer is only the ones a human wrote down in advance and tested against history. A rule that would have fired forty times last month is not an alert, it is a feed. Replaying it before arming it is the only way to know.
Then there is the failure nobody plans for: the send itself fails. A transient error from Slack marks the channel broken, the tool stops trying, and every alert after that silently goes nowhere. The right behaviour is to retry on a short ladder, count consecutive failures before giving up, and show a resume button when it does.
InternalStatus ships those defaults: severity thresholds per channel, quiet hours that critical alerts ignore, a retry ladder, and a backtest before any rule can arm. The result is a channel where every message earned its place.