Fwiw, removing ALL_TRUSTED from the shortcircuiting meta definition certainly prevents it from triggering the shortcircuit.
However, it still fires -- incorrectly & intermittently, albeit now with a "-1" score attached. The LHLO/LMTP header still is added at the backend, and UNPARSEABLE_RELAY still hits. I've not yet determined what the actual problem with the parsing is.
