Live runtime stats for the operator console's dashboard.

What an operator wants during an incident, in one cheap read: is the node healthy, how much is it holding, and how many players are on it. Everything here comes from the VM or from asobi_presence - nothing touches Postgres, so this endpoint stays answerable when the database is the thing that is unwell.

Why this is polled rather than pushed

The equivalent in the old asobi_admin was a WebSocket that pushed every two seconds, and it carried its own in-protocol authentication because a ws upgrade cannot be rejected cleanly before it completes. The ops plane already has bearer tokens and capability classes; a second auth path for a payload this small is not worth maintaining. The console polls.

Why node is reported

asobi clusters (see the clustering guide), and every node serves its own copy of this endpoint. An operator behind a load balancer would otherwise have no way to tell which node answered - and since the numbers here are per-node, a reading without a node name is a reading you cannot act on.