Deep research gathers an incomplete orchestrion location dataset
Willison ran o4-mini-deep-research to find surviving orchestrions and return structured location records. The model used web search and code execution; he explicitly considered the resulting list incomplete and unverified.
til.simonwillison.netWhat happened
Willison reports a JSON dataset of locations, but rejects treating it as comprehensive or publishable without fact-checking.
Reported by Simon WillisonHow the agent did it
- Submit a structured location-research request through the background Responses API.
- Let the model search websites and use code interpreter for supporting calculations.
- Poll for completion and inspect the JSON records and tool trace before trusting the dataset.
Where a person stepped in
Willison inspected the trace and warned that the returned facts still needed verification.
A person approved each step before the agent continued.
More details
About this account
The research model chose and executed multiple searches, page visits and code operations after one research request.
Firsthand account links the full API result and trace. The reported research output is not a verified directory of instruments.
Credit: Simon Willison
Added 14 Oct 2026
Source published 18 Oct 2025
Source checked 14 Oct 2026
Tools and topics
o4-mini-deep-research · Responses API · Web search · Code interpreter · curl · jq
Research · Sources
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.
Prompt to try
Editorial adaptation, not the source author’s exact prompt. Replace bracketed inputs and review tool permissions first.
Research [bounded set of public objects] and return JSON with name, location, primary source URL and verification notes. Record the search scope and exclusions. Do not claim completeness; distinguish source-confirmed facts from estimates and provide a queue for human fact-checking.