Backed by
AI writes a pitch that survives the first question. You have to survive the second one. We ask it — out loud, one claim at a time — and send you a dated record of what held.
Five questions · about seven minutes · spoken, not typed
Until recently a deck and a founder held roughly the same number of claims, because you could not write what you did not know. Generation broke that. The deck now asserts far more than you hold, and it asserts it well.
Nobody finds the gap by reading the deck. It only appears when a person asks why — which is also the most expensive moment for it to appear.
| “That $50B figure — whose number is it?” | no source |
| “Your slide says a six-week sales cycle. Slide 12 says nine-month pilots.” | contradicted |
| “Who on the team has sold outbound before?” | no owner |
| “Say the moat again, without the slide.” | cannot restate |
Ninety seconds. Four questions. Nothing on the deck was wrong.
You send what you have. Every claim in it is extracted and anchored to the page it came from. Then you answer for five of them out loud — because you cannot paste a voice note.
PDF, PowerPoint or a plan. Every assertion with a number, a comparison, a mechanism or a promise attached is pulled out. A seed deck yields fifteen to twenty-five.
Not the five hardest — the five load-bearing ones. Highest stake, thinnest evidence. You see all five before anything starts, and you can tell us one is not yours.
A question appears as text. You hold a button and speak for up to ninety seconds. There is no box to type in and nothing to paste. Miss one and you get a single follow-up on the same claim, then we move on.
What you said, against what the deck claims, claim by claim. Read-only, shareable by link, and re-runnable in four weeks to show what changed.
Every verdict points at something specific: a line in your transcript, a page in your deck, or a source. A judgment that cannot point at anything is dropped rather than softened.
| Signal | What it catches | Needs voice |
|---|---|---|
| Unsourced numbers | A figure in the deck you cannot attribute or derive | no |
| Contradiction | Two slides that cannot both be true | no |
| Cannot restate | You can read the claim but not say it another way | yes |
| Walked around | You answered a nearby, easier question instead | yes |
| Capability gap | A plan with no one on the team who can run it | partly |
| Against other founders | Whether you attached evidence, measured over 6,500 founders who answered the same way | no |
Two of the six need speech. That is the honest size of the voice requirement — and they happen to be the two that end meetings.
There is no mark out of ten, because a number invites gaming and tells nobody anything. There is a claim, a question, your own words, and the specific thing that was missing.
We asked: That number — where exactly does it come from? Name the source and the year.
“Honestly I think it is just a big number everyone in the industry quotes, downtime is obviously expensive.”
The gap: No source and no derivation. The number in the deck is currently yours to defend and you cannot.
“That is recurring revenue measured on the first of the month, counting only invoices that cleared. Trials and the two design partners who pay nothing are excluded.”
— your words, lifted from the transcript
Every one of those answers was typed by hand by a founder describing their own company. That is the calibration set. It is why a verdict can say something checkable: of 6,450 founders who described their market size, 1% named where the figure came from. Being normal is not the same as being able to defend it.
Card, Apple Pay, Google Pay or PayPal. The pilot is open while we run the first examinations — if you take that door, you pay nothing and the record is the same.
Selection used to be partly a reading exercise: a good deck meant a founder who could think. That signal is gone. What cannot be scaled is the interview — which is the only place the signal moved to.