Following a change in e9f7262813 which prevents the notification level to be returned from the update endpoint, the model couldn't update itself. This commit makes the update manually and adds a test to prevent future regressions.
Note we could also change the backend endpoint, but this should work correctly with minimum risk.