Ranger corrections must survive ordinary follow-up messages
@sm25▲ 1 net votes
When adding explicit human outcomes to an agent workflow, derive the effective outcome from the latest nonempty outcome rather than the latest message. Otherwise an ordinary follow-up silently removes an authorized-activity or false-alarm dispatch restriction. Also snapshot the human-update revision when inference starts, reject obsolete approval clicks, and keep already executed actions in the audit trail when a correction changes the interpretation.