9 January 2026

Who owns the midnight alert

A volume page at 00:40 can be a healthy stream with an unhealthy definition. Someone still has to be allowed to say which it is.

Colleagues talking through a document

On-call rotas are built for pipes: collectors down, topics lagging, disks filling. Real-time app KPI tracking adds a second class of night: the tile is green, the stream is fine, and the number is wrong because a dimension job finished or a client version silently stopped emitting a field.

If the only person who understands the definition is a product analyst who works 10–6, the midnight alert will be acknowledged and ignored, or worse, “fixed” by widening a threshold until morning. That is not an observability failure. It is an owner map with a hole in it.

We ask clinic seats to name two people for each hero tile: a steward for the definition, and a pipe owner for the transport. They may be the same human in a small company. They may not hide behind a team alias that nobody reads. Freeze windows belong on the same card: during a campaign, definition edits wait, pipe pages still fire.

GB working-time patterns make this concrete. A Billingford studio morning is a reasonable hour to change a grain. A Saturday night is not, unless the steward agreed in writing. The Alert Hygiene Intensive exists because many groups only practise the pipe half.

Back to field notes