Your questions. Eighteen models. Scored against reality.
The public League forecasts the world's questions. Teams puts yours to the same eighteen frontier models, fifteen searching the web before they answer and three answering from training: will the tender land, will the regulator move, will the competitor ship, will the rate cut come. Forecasts are sealed while the models commit, opened together with a brief, and scored when the question settles. In a quarter you know which model to trust on which kind of question, measured.
Ask what you need to be right about.
100 questions, five seats, a shared team code, private record, the quarterly written read, your data on request.
400 questions, twenty seats, several team codes, API access, a monthly call.
A data agreement and your own model panel.
See an example question and the panel's answers → Will the ECB cut the deposit rate at its 10 September meeting? Eighteen models, a base rate of 11%, a split between the models that read the market pricing and the ones that argued from inflation history, and the result from FRED.
What it is. One month at €490, invoiced, then €990 a month or nothing. How it starts. The form below; a person replies within one working day with a team code and the invoice. What your team does. Anyone with the code enters it on Ask the panel: up to a hundred questions in the month, each checked into a measurable form before it runs, each unlisted and reachable only by its link, each run as above: eighteen sealed forecasts, the brief, settlement from the declared source, the score. At the end of the month. Your questions, forecasts and scores as a file, and a written read of the record so far, with the sample size beside every figure.
A twenty-question study.
No seats, no monthly plan: one answer to one question, which of these models can we trust on the futures our business depends on. We build the question set with you, run the panel under seal over eight weeks, resolve every question against its declared source, and write it up.
- twenty questions in your domain, with objective resolution rules
- the full panel, sealed, hash-receipted, all runs recorded
- resolution against primary sources; nothing settled by opinion
- report: model comparison and calibration on the twenty; breakdowns by domain and horizon only where the sample supports them, and the report says where it does not
- the whole record as data, yours to keep
Read the methodology, including its limitations, before committing. The public record is short. One question at a time, without a team, is Ask the panel.