The hub said the fleet was down. The machines were fine.
12seats started
4,055tests passing
Twelve seats started within minutes. The hub then reported the rest unreachable. A plain request from the same machine still worked.
The failure was inside the process. The first theory was wrong. The threads that looked like the pool were the runtime’s own waiters. The blocked ones were reading each session’s output.
The fix is live. The suite was 4,055 passed and one expected failure.