Don't act on advice nobody checked.
Three AI advisors argue your decision out loud. Every factual claim they make is checked against live sources, so you can see which parts hold up before you act on any of it.
the sample runs are recorded — they start instantly
Roughly 90% of startups fail, so the equity is worth nothing.
Confident is not the same as correct.
You get a single confident answer. Nothing tells you which parts are true.
Now you get five confident answers. It feels more rigorous and is no more reliable.
The debate happens, then every claim in it is checked, and you see which ones survived.
- 01
It frames the decision
Restates what you are actually choosing between, and asks the three questions only you can answer.
- 02
Three advisors argue it
An Advocate, a Skeptic and a Pragmatist take two rounds. They are told to make checkable claims, not to sound wise.
- 03
Every claim gets checked
Each claim is pulled out, tagged, and searched against live sources. A refuted claim forces its author to retract in round two.
Most tools collapse everything into true or false. The two most useful categories are the ones in the middle: no evidence found is not the same as false, and depends on facts only you know is not a failure of the check — it is the question you still have to answer.
- VerifiedIndependent sources support this claim.“About a third of seed companies raise a Series A.”
- DisputedSources conflict with this claim, or contradict it outright.“Roughly 90% of startups fail.”
- UnsupportedCheckable in principle, but no evidence was found. Not the same as false.“Engineers who return avoid a title downgrade.”
- Depends on youTruth depends on facts only you know about your situation.“This assumes your savings absorb a 30% pay cut.”
- OpinionA value judgment. Not the kind of thing that can be checked.“Take it only if you'd take it at zero equity.”
An honest ledger
“This recommendation rests on 6 verified and 12 unverified claims.” Nobody else shows you that number.
A scoreboard
Which advisor's claims actually survived checking — and which ones they were forced to withdraw.
Open questions
Every claim that depends on your situation becomes a question, already tied to the argument that needs it.
Something to do
Two or three concrete next steps, plus the single finding that would flip the recommendation.
Debate and synthesis on Groq · verification against live web search · roughly 13 model calls and 13 searches a run
Counterpoint does not determine truth. It separates what has been checked from what has not, and shows you which is which.