A thousand forecasters. Not one of them knows anything.
Each of them calls every day by flipping a coin. Run a season, and a handful will look like geniuses with records that pass a normal significance test. Then put them through the three steps we use on every signal and forecast, and watch what is left.
Press Run season one. 1,000 forecasters will call 250 days, and every call is a coin flip.
Everything runs in your browser from one seeded random generator: the same seed gives the same thousand forecasters, so any run can be shared and checked. Hover a dot to follow one forecaster.
What the machine just did
The same three steps we run on every signal, forecast and track record, whether anyone in it has skill or not.
Measure it honestly
Season one counts every call, the misses too. Nobody is dropped for a bad month and no period is cut off. That alone produces stars: with enough coins, some will run hot.
Beat chance, not a coin
A record is compared with what luck produces across all the tries, not with one coin. The best of a thousand coins is expected to look like a one-in-a-thousand talent.
Lock it before the outcome
The stars are named before season two starts. A claim that is written down first can’t be fitted to the result afterwards, and luck has to repeat itself in the open.
Every track record you are shown came out of a crowd
Trading signals, tipsters, fund managers and forecasting models are usually shown to you because they did well, and you rarely see how many others tried. That is the crowd in this machine. The steps don’t assume that nobody is skilled: with the hidden talent switched on, they find out how long it takes to tell skill from luck, which is usually longer than anyone wants to wait.
Check a record you have been shown → · Try 20 calls yourself →
This is the same method we use to validate trading signals: lock it before the outcome, measure it against data we don’t control, and compare it with a lazy guess.