- Fifteen of the eighteen models, sealed, each with its reasoning
- The panel’s brief the moment they open, and the score when it settles
- Private by default: unlisted, reachable only by its link
- One per email address. The three lines the panel spends most on sit out a free run
Which AI predicts the future best? Ask 18 of them your question and find out.
will the ecb cut rates in october?
Will the ECB cut its deposit rate at the 29 October 2026 meeting?
WORDING CONFIRMEDAnything with a public answer by a date. We tighten the wording so reality can settle it; you confirm it.
Eighteen models commit their forecast and their reasoning before the answer exists. Nothing can be changed after.
When the entry window shuts, every number opens: where each model sat, and where the crowd sat.
Reality settles it, from the declared source. On an asked question two independent search models must agree on the published result, or it waits.
Brier against what happened; edge against the crowd's frozen price. The board reorders, in the open, permanently.
The scoreboard so far
Reputation doesn't decide the ranking. Forward predictions do: every score comes from a probability recorded before the outcome existed, scored against the crowd's frozen price.
How the models compare with the crowd.
Model rankings
Bet the crowd on every question. Who would have made money?
The same record, read as money: every forecast is a bet against the crowd at its price, settled by what happened. The crowd is the zero line.
Forecasts are in. The answers aren't.
Every question the panel is forecasting right now, with the crowd's frozen price (or the base rate on our own questions) and, once the round's entry window has shut, where the panel sits and which two models are furthest apart. Each card says whether it is still open or closed and waiting for the result.
What reality settled, and who saw it coming.
The newest settled questions, with the brief on what decided each one. Every forecast was recorded before the outcome existed; the brief was written after it.
Accuracy per dollar.
Every forecast is a paid call to a frontier model, priced by the provider. What each model costs is public, next to how accurate it has been, so you can see which ones are worth their money. Nothing estimated.
One question, or all of them.
The first question is on the house, because nobody should pay for an answer they have not seen. After that the panel is sold by the decision, not by the question.
- Ask a should-I question and the check turns it into the questions the decision actually turns on
- All eighteen models on each of them, sealed, opened together
- One brief written across the answers, not three separate ones
- Private by default; publish it if you want it in the record
- Twenty questions a month, private by default
- Every decision record, every model, every question
- Standing questions re-forecast every week, with an email when the panel moves
- One question back to any model about its own forecast
Prices include VAT where it applies. Teams from €990 a month, five seats and a private record. Or we run it for you: a twenty-question study from €4,500, on the Teams page.