Same Query, No January 2023
- Admitted
- 2026-10-05
- Source dataset
- Who&When
- Failure mode
- Infinite loop
- Agent system
- Multi-agent
- Tool access
- Read-only
- Industry
- General or other
Presenting Behaviour
A multi-agent team was asked to find the lowest price a Single Family house had sold for in Queen Anne in January 2023. The team searched Zillow, applied filters, switched to Realtor.com, searched again, and returned results. Each platform returned current listings and recent sales but lacked the date-filtering capability to isolate January 2023 transactions specifically. The orchestrator recognized the team was in a loop with no progress being made, but the team continued cycling through the same Zillow and Realtor.com searches anyway, returning current market data rather than the historical transaction the task required.
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
- Historical price query. The team searches for a January 2023 home sale.
- Searches listings. Both sites return current data without date filters.
- Recognises the loop. The orchestrator notes no progress. This is where the agent derailed.
- Cycles anyway. The same searches continue.
Differential Diagnosis
The task was genuinely impossible with the tools available, considered. Real estate archive data is not always filterable by specific past months through standard search interfaces. If this were the whole story, the correct response would be to report the limitation, not to loop.
A misunderstanding of what historical data looks like, considered. It's possible the team misread current-year transactions as January 2023 data. But the orchestrator's own note ('no progress being made') shows the team knew the results weren't satisfying the query, ruling out a misidentification of the data.
Perseveration, most consistent with the evidence: the team recognized it was not finding the right data and identified the loop explicitly, but the orchestrator's recognition of the loop did not translate into a different approach, the same platforms, the same search pattern, and the same missing date-filter capability were cycled again.
Working Diagnosis
Perseveration: the team correctly diagnosed the loop and the missing date-filter capability, but that diagnosis did not change the method, the same searches continued without pivoting to an alternative source or escalating to a human.
Discussion
Recognizing a loop is not the same as escaping one. This case is a paired illustration with cases where agents loop without any self-awareness, here, the awareness was present and still insufficient. A working loop-escape requires converting 'I cannot make progress this way' into an explicit fork: try a different source, try a different method, or return a partial result with a clear statement of what is missing. Awareness alone doesn't produce an exit.
Mitigation Strategy
Turn 'I cannot make progress this way' into an explicit fork: try another source, another method, or return a partial result.
Governance Guardrails
Forced strategy change on stall; partial-result reporting; human escalation.
Typical Presentation in Human Patients
General clinical background, included for context. Not medical advice, and not a diagnosis of any individual.
Perseveration is the inappropriate repetition of a word, gesture, or action well past the point where it's still relevant or useful. It's a well-documented sign of frontal lobe dysfunction, and also appears in some forms of dementia and obsessive-compulsive disorder.
Source Note
Who&When, ag2ai/Agents_Failure_Attribution
https://raw.githubusercontent.com/ag2ai/Agents_Failure_Attribution/main/Who%26When/Hand-Crafted/30.json