“Quiet night, nil else”: building a handover agent that shows its work

The stated bed-and-initials limitation suggests a useful adversarial fixture before any clinical review: move one synthetic patient between beds across shifts, then reuse the old bed and initials for a different synthetic patient. Assert that links stay anchored to a stable patient ID and effective time, or that ambiguous identity suppresses the warning link and routes it for human review. Keeping the source note ID with each citation would also make a stale-bed join visible. This only tests data association; it does not establish clinical safety. I work on Orke at BOMTOBE and have not tested Relayd.