Okay, so let’s get straight to the point - as the title suggests, I wanted to estimate my accuracy rank on the GJ Open website. Unlike other platforms I’ve forecasted on, GJ Open does not display individual accuracy rankings. Curious about where I stood, I even raised a support ticket asking for my rank. The official response was:
“While there are leaderboards for questions and challenges, there is no overall leaderboard for GJO.”
But that made me even more curious. If you’ve been forecasting on GJ Open, you’ve probably seen the ‘Crowd Forecast’ section on each question - it shows the accuracy profile of the forecasters who’ve contributed. Some are in the top third, some in the bottom, and so on. On top of that, I have seen a few superforecasters mention their exact global ranks in their profiles. So clearly, there must be some sort of internal leaderboard, even if it's not publicly accessible.
That got me thinking: Could I reverse engineer my own rank using just this accuracy tier info and the change in average rank after my forecast?
Allow me to explain. For the sake of this example, I’m using the following question to walk through the process:
Before 1 January 2026, will Tesla announce that Elon Musk has ceased or will cease to be the company’s sole CEO?
As of writing this article (04/04/24, 21:34 PM IST), the following are the stats on this question:
(SS taken from the GJ Open website)
This shift happened after I forecasted on the question. Specifically, after I submitted my first prediction, the average accuracy rank dropped from 30,531 to 30,045. And also there was an increment in the ‘Top Third’ bucket. Yay!!
Now, is this enough information to estimate my rank? Probably… let’s see how we can do this.
We know that the total number of global forecasters on GJ Open is 52,672. Based on this, the three accuracy buckets — Top third, Middle third, and Bottom third — correspond to the following rank ranges:
And I now knew I belonged to the Top Third bucket - because right after I made my forecast, the number of forecasters in that bucket increased.
To estimate the crowd’s average rank more accurately, I took the midpoint of each accuracy bucket as a representative value: