How judging works
Every score points at the second it rests on.
AI judges are fast and tireless, and they can also be confidently wrong. So ours work under four rules that are enforced in code rather than promised in a policy, humans have the final say, and every founder can see exactly why they got each score.
The four rules
Each rule has a name, because each one is a real check that runs. This is what it says, and how it is held.
rubric_onlyJudges score only against the published criteria and cite evidence from the pitch.
A score that names a criterion outside the rubric, has no timestamp, or quotes words that aren’t in the transcript is rejected before it is saved.
no_conflictsA sponsor’s judge can’t score a founder they’re connected to.
A connected judge is taken off the pitch before seeing its scores, and the database refuses a score or a confirmation from them.
human_final_sayAI scores advise. A human judge confirms every winner before anything is paid out.
Only a human judge on the panel can confirm winners, and only once every pitch is scored and no two winners are level. The prize ledger issues nothing before that.
prize_scopeThe payout agent can only issue the prizes that challenge holds. No improvising.
The payout holds a key for one challenge. It can issue that challenge’s prizes to confirmed winners, once each, and it never sees a prize code.
From pitch to prize
The rubric is published
Criteria and weights are on the challenge page before anyone applies. They don’t change afterwards.
The pitch is transcribed
What is said is written down with timestamps as it happens. That transcript is the evidence.
Each judge scores each criterion
From 0 to 10, with the timestamp the score rests on, the words said then, and a sentence of why. Two AI judges and the human judges all do the same.
Totals are worked out
Each judge’s scores are weighted by the rubric into a total out of 100. A startup’s total is the average across the judges who scored it.
A human confirms the winners
When every pitch is scored and no two winners are level, a human judge confirms the result. Then the prizes go out.
Founders get the replay
Every score, who gave it, and the moment it cites. Any score can be appealed to a human who didn’t score that pitch.
What the AI judges can and can’t do
There are two: a tech judge, which weighs what was built and shown working, and a market judge, which weighs the problem, the customer and the evidence that people want it. Each scores one criterion at a time and sees only the transcript.
They can’t add criteria, can’t score without citing the pitch, can’t see who a founder is beyond what was said on stage, and can’t decide who wins. Their scores sit beside the human judges’ as a second opinion, in blue, so they are never mistaken for a person’s.
Questions
No. AI scores are advice. A human judge confirms every winner, and nothing is paid out before that.
Every score must quote words from the pitch and give their timestamp. The quote is checked against the transcript, and a score whose quote was never said is rejected.
Founders declare connections when they apply, and judges can flag them on the day. A connected judge is taken off that pitch before seeing its scores.
Yes. Any score can be appealed from the replay page. A human judge who did not score that pitch reviews the cited moment and rules on it.
There is no tiebreak. Winners cannot be confirmed while two startups inside the winning places share a total, so the judges score again until they differ.