Every finding we propose goes through the same step before anyone writes a word: we go looking for the person who already published it.
Search, then the communities where the question actually gets asked, then the AI assistants — because if an assistant answers the question directly and cites a source, the finding is not unclaimed, it is already someone else’s.
What it looks like when it works
On a recent run we proposed three findings for an infrastructure company. Two came through clean, with nothing comparable in public. The third was about how job listings describe infrastructure security responsibilities — a good idea, well framed, and one that would have made a genuinely interesting piece.
It did not survive. A UK site has been publishing exactly that data continuously for years, broken down by role, updated monthly. Worse than being second: the public data contradicted the framing we had built.
Why we show you the ones we killed
Three ideas is a blog post. Three ideas with the reasoning behind the ones that died is an argument — it is the difference between a list you have to trust and a process you can check.
It is also the answer to the obvious question, why can’t I just run deep research myself? You can. You just cannot run it on everything, and the value is not in the one idea that survives. It is in knowing that forty others did not.