Everyone has opinions. You have a percentile.

By comparing what you announce to what happens, across a large number of decisions. Percentyl applies the method of forecasting tournaments: you put a number on a real case, you lock it before the outcome is known, and reality settles it. After thirty resolved cases, two things become measurable: your accuracy against the crowd, and your calibration, meaning the gap between the confidence you state and how often you turn out to be right.

A decision maker is well calibrated when the things they announce at 70 % happen about 70 % of the time. It has nothing to do with being right often: you can be very knowledgeable and very badly calibrated, saying 95 % for things that happen two times out of three. Calibration is the rarest and the most transferable judgment skill, because it bears on how you dose your confidence, not on a field.

It is not being against the crowd on principle, which still amounts to letting the crowd decide. It is knowing when the consensus sees right, and when it is wrong. Peter Thiel puts it in a single question: what important truth do very few people agree with you on? A useful contrarian follows the majority most of the time, and departs from it at the precise moment it is wrong, with reasons that can be checked before the outcome, not told afterwards.

On knowledge and speed, by far. But a language model gives back the consensus of what has been written: massively reasonable, rarely contrarian at the right moment. And the value of a decision plays out mostly when common thinking is wrong. That is exactly the ground Percentyl measures, and that is why the ability to depart from the consensus with good reasons becomes more valuable as reasoning becomes free.

On each case, your error is compared to the median player's: you gain or lose 25 points at most, never more. Your displayed score is the average of those gaps, so it does not depend on how many cases you have played: playing more does not inflate the score, it makes it reliable. The method adds a bounded bonus, and nothing is public before thirty resolved cases, all fields combined.

Yes, and it has been measured. In the geopolitical forecasting tournaments of the 2010s, one hour of calibration training durably reduced participants' error, and the performance of the same forecasters persisted from one year to the next: it was not luck. Philip Tetlock identifies the best predictor not as intelligence, but as the habit of revising your beliefs, which weighs about three times more.

No, and it is structural: no stakes, no purchasable currency, no cash winnings, no investment advice. Points and percentiles are fictional units that cannot be transferred. What you earn here is a score that can be checked, and the right to bring it up at lunch.

Probably at first, like almost everyone: overconfidence is the best documented bias in the literature, and discovering yours as a number is precisely the information worth gold. Your score stays private before thirty resolved cases, time enough to learn without an audience. After that, you decide what you show.

The app is free and will stay free: one case a day, the contrarian thinking course included, the leaderboard and the public profile. Premium services will come later, around the community. No ads, no data selling, no financial advice.