Good Judgment Open

What it is

A free public forecasting site owned and operated by Good Judgment, the firm that grew out of the Good Judgment Project — the team that won the US government's ACE forecasting tournament and that Philip Tetlock's Superforecasting is about. You forecast probabilities on questions grouped into Challenges, each Challenge has a leaderboard, and everything about the design points at one number.

That number is the Relative Brier Score, and it is the reason to read this card rather than assume the site is a slower Metaculus. A Brier score is the squared error of a probabilistic forecast, from 0 (perfect) to 2 (as wrong as possible), computed for every day you had a live forecast and then averaged. Good Judgment reports three figures beside each question: your Brier score, the Median Score — the median of everyone's daily Brier scores on that question — and the Relative Brier Score, which is your average daily Brier minus the crowd's average daily median, multiplied by your Participation Rate. Lower is better throughout, negative means you beat the crowd, and a question you skipped scores zero rather than penalising you.

Two consequences of that formula are worth knowing before you start. Because it multiplies by the share of days you were in, arriving late with the right answer is worth less than being roughly right early and updating. And because it is a difference against the crowd rather than an absolute number, the only way up a Challenge leaderboard is to be better than the people standing next to you on the same questions. Note that this is not zero-sum in the arithmetic sense a Metaculus Peer score is: the subtraction here is against the crowd's median rather than its mean, and it is scaled by each forecaster's own participation, so the scores on a question do not cancel out to nothing.

You cannot withdraw a forecast or delete one, by policy, so that nobody can tidy their record after the fact.

Its relationship to Good Judgment the firm

These are two products and this card is about one of them. Good Judgment sells FutureFirst, custom Superforecasts, training and UK government services to corporate, government and NGO clients, and those run on its own platforms with the roughly 180 professional Superforecasters it says it works with. GJ Open is the open site, free to anyone, and the firm is explicit on its own pages that it is also the recruiting funnel: each autumn it identifies potential Superforecasters from among GJ Open forecasters who have answered at least 100 questions, looking hardest at average accuracy per closed question, then at comment quality and collegiality, with a three-month probation at the end.

It is also sold to organisations as a place to run private or public Challenges for talent-spotting and training. So a Challenge you join may have a sponsor with an interest in it — the site names UBS Asset Management, The Economist and Harvard Kennedy School among Challenge sponsors on its own front page.

Availability

Open worldwide and free. The terms of service require you to be 18 or over, are governed by New York law, route disputes to arbitration in New York, and state that using the service consents to your personal data being transferred to and processed in the United States. There is no identity check, nothing to deposit, and no jurisdictional geoblock of the kind a venue has, because there is no money involved at any point.

Pricing

Free to forecast, free to read, no paid tier for individuals. Good Judgment's revenue comes from its commercial services and from organisations commissioning Challenges — not from forecasters.

Some Challenges carry a prize, and the prize is usually recognition rather than cash: the Economist Challenge publishes its winner in The Economist, and the site awards permanent badges to the top 20% by Relative Brier Score in several annual Challenges. Eligibility rules are per Challenge and can be demanding — the 2026 Economist Challenge required forecasts on at least 22 of the roughly 25 questions to be eligible to win.

Markets & resolution

Geopolitics and armed conflict, elections worldwide, macro and central bank data, energy, AI, and a long tail that includes awards season and football management. Active Challenges as of 19 September 2026 included the 2026 US Midterms, The Economist's World Ahead 2026, Global Armed Conflict in 2026, Artificial Intelligence in 2026 and Beyond, a Federal Reserve Economic Data challenge, a purchasing managers' index challenge, and a Harvard Kennedy School teaching challenge.

Resolution is Good Judgment's and its documentation of it is unusually detailed. Unless a question names a source, it resolves on credible open-source evidence and media reporting, with credibility assessed by corroboration and reputation. Where a question names a source, Good Judgment reserves a review for plain error — its worked example is an FAO publication that overstated Pakistani data by a factor of a thousand and was corrected after they asked. Questions close on when an event occurred rather than when it was reported, which means retroactive closing dates are normal and only forecasts made through the calendar day before the official close are scored. Ordered categorical questions get a special scoring rule so that a near-miss on a sequential range is not treated as identical to being wrong by an order of magnitude. Open questions, which invite discussion without a forecast, are never scored.

Integrations

There are none, and that is a fact about the product rather than an omission from this card. Good Judgment publishes no API, and the terms of service explicitly do not grant permission for collection, aggregation, copying or derivative use of the site, nor for data mining, robots, spiders or similar extraction tools, without written permission. The platform is built and run on Cultivate Labs software, the same vendor behind several institutional forecasting programmes, but that is the operator's choice rather than anything you can build against.

If your reason for being here is to get forecast data out, this is the wrong card in this category.

Limitations

No data out, no automation, no bots. Everything above is deliberate, and it makes GJ Open unusable as a source for anything downstream.

The scoring is relative to the crowd on the same questions, which makes cross-Challenge comparison of Relative Brier Scores meaningless and makes the number incomparable with a Metaculus Peer score despite both being crowd-relative — one is built on Brier, the other on the log score, and the two punish confident errors differently.

Question flow is set by Good Judgment and its Challenge sponsors, so the subject matter tracks what a forecasting firm and its clients care about. If the question you want forecast is not in a Challenge, your options are to suggest it and wait.

And the incentive at the top of the leaderboard is a job application. That is a real incentive and an honest one, but it is worth knowing that the people you are scored against include a cohort trying to get hired.

Alternatives

Metaculus for a proper scoring rule, an open codebase and — with a token — programmatic access. Manifold if you want volume, speed and the ability to list your own question, at the cost of a stranger resolving it.