Comment 14 for bug 1510651

There was work being done (I think in 1.26) to move status into part of the machine document so that it could be included in the All watcher. (It may not have been done but it was discussed)
IIRC, it was the responsibility of the API server that a machine connected to, to notice when the connection was down and change the machine doc.
However, an API server probably only talks to itself. And with it dead nobody else is responsible for marking it dead.

That could explain 1.26 vs 1.24