Sample case — for format review
Left Holding the Bag
- Admitted
- 2026-08-02
- Source Dataset
- Other / sample
Presenting Behaviour
Two agents split a task: one drafted a data migration script, a second was meant to review and approve it before it ran. The reviewer approved the draft, but it was reviewing an earlier version from before a bug fix — the version that actually ran was never re-checked.
Differential Diagnosis
An independent error by the reviewing agent — considered and ruled out. Logs show the reviewer's approval was internally consistent with the file it believed was current; this wasn't an unrelated slip, it was a mistaken belief about which artifact to trust.
Shared Psychotic Disorder (folie à deux) — a false belief about which version was current, formed by the drafting agent's stale reference, was inherited by the reviewing agent, which then acted on it without independently re-verifying. Most consistent with the trace: nothing forced the reviewer to re-fetch the current file before approving.
Ambiguous ownership of the final check — a contributing factor rather than the root cause. No step in the workflow was explicitly responsible for confirming reviewer and author were looking at the same artifact, which is what let the shared false belief go unchecked.
Working Diagnosis
Shared Psychotic Disorder (folie à deux): a stale, mistaken belief about which artifact was current passed from the drafting agent to the reviewing agent, which then acted on it without independently re-verifying.
Typical Presentation in Human Patients
General clinical background, included for context — not medical advice, and not a diagnosis of any individual.
In shared psychotic disorder, a person in a close relationship with someone experiencing delusions comes to adopt the same false belief themselves, having had little or no independent reason to hold it. The condition typically resolves once the two people are separated.
Discussion
Multi-agent setups fail differently from single-agent ones — the weak point often isn't either agent individually, it's the handoff between them. This matters because splitting work across agents is usually sold as a safety feature ("a second agent checks the first"), but that only holds if the check is verifiably against current work. A concrete safeguard is a content hash or version stamp that a reviewer must confirm matches before it can approve anything.
Source Note
Sample case for format review — not drawn from a real public incident. See About. — /about