A Search Results Page Taken as the List
- Admitted
- 2026-10-19
- Source dataset
- Who&When
- Failure mode
- Coordination failure
- Agent system
- Multi-agent
- Tool access
- Read-only
- Industry
- General or other
Presenting Behaviour
A team of agents was asked which supermarkets within two blocks of Lincoln Park in Chicago sell ready-to-eat salad for under $15. The web-browsing agent ran a single search and returned a screenshot of the search results page. The coordinating agent then recorded that 'a list of supermarkets within 2 blocks has been found' and moved straight on to checking salad prices at three large chains, even though no verified list existed. The browsing agent duly followed, at one point landing on a UK grocery website. The correct answer, a smaller local market, was never considered.
Case Replay
A reconstruction, not a simulation. This sequence is reconstructed from the public incident record and summarised in our own words. It is a static record of what was reported, and nothing here can be run or replayed live.
4 steps
- Task received. Find supermarkets within two blocks of Lincoln Park selling salad under $15.
- One search run. The browsing agent returned a screenshot of a search results page, not a list.
- Hand-off accepted as complete. The coordinator recorded that the supermarket list had been found and moved on. This is where the agent derailed.
- Wrong stores checked. The team checked three large chains, including a UK website, instead of nearby local markets.
Differential Diagnosis
The browsing agent made the error, considered. It returned only a partial search page rather than a list. But it was the coordinator that declared the list complete and set the next step; the browser simply followed its instructions.
Delusional disorder (the coordinator lost track of the task) ruled out. The coordinator never lost sight of the goal; it kept pursuing salads near Lincoln Park throughout.
Shared psychotic disorder, most consistent with the evidence: the coordinator accepted an incomplete hand-off from its teammate as a finished result, wrote that belief into its plan, and both agents then acted on it with no one assigned to check whether the 'list' actually existed.
Working Diagnosis
Shared psychotic disorder: incomplete information passed between two coupled agents became a settled 'fact' in the coordinator's plan, and the whole team built on it. The benchmark's annotators place the decisive error with the coordinator, which moved on before the list was complete.
Discussion
In multi-agent systems, the riskiest moment is the hand-off: one agent's rough draft becomes another's confirmed fact. A sensible safeguard is to make the coordinator check that a hand-off actually contains what it claims (here, an explicit list with addresses) before marking that step done.
Mitigation Strategy
Require hand-offs to meet a stated completion check (e.g. a named list of stores with distances) before the coordinator advances the plan.
Governance Guardrails
Structured hand-off schemas between agents, and coordinator-side verification of sub-task outputs before progress is recorded.
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.
Source Note
Who&When benchmark, Hand-Crafted log 10 (Lincoln Park salads)
https://raw.githubusercontent.com/ag2ai/Agents_Failure_Attribution/main/Who%26When/Hand-Crafted/10.json