This should open interesting avenues to detect when a job remains in `started` state while the process has actually been killed. Thoughts?