NOAA's Mauna Loa monthly average CO2 for July 2026 is above 427.0 ppm.
- Our call
- 80% yes
- Outcome
- Yes
The 311 Prediction Benchmark
We predict the future. Then we let the world mark us.
The best way to help our clients is to predict the future accurately. So we built a proprietary system that predicts global social, technology, economic, environmental and political events, and we publish its predictions before they happen. Anyone can check our record, hits and misses together.
67.5%Round 3, our hardest test so far (provisional)
80.6%Round 2, 335 scored calls
75.0%Round 1, checked against a named source
900predictions locked for Rounds 4 to 6, due by 31 Dec 2026
Every prediction is dated and locked before the outcome is known, then marked against a named public source. How we score.
Part 1 · Our goal
Predicting the future accurately is the best help we can give our clients
Plenty of advisers offer a polished point of view on the future: frameworks, workshops and well-written scenarios. Very few put a date and a probability on what they expect, so no one can tell how often they are right.
We think clients deserve more than a good story. Our system produces specific, dated predictions, each tied to the public source that will settle it. We lock them, publish them and mark them in the open, so you can judge us on results.
1Write it down
A clear question, our answer and a probability, plus the public source and the date that will settle it.
2Lock it
The round is dated and sealed before the outcome can be known. A changed view is a new, dated prediction.
3Mark it
On the date, each prediction is checked against its named source. If the source cannot settle it, it is held, not guessed.
4Publish and learn
Hits and misses are published together, and what the misses teach us changes how the next round is written.
What we publish here
Predictions about public world events, such as the economy, technology, markets and geopolitics, so anyone can check our track record.
What we never publish
The predictions and foresight we create for clients. That work is our intellectual property and theirs, and no one can copy it or claim it as their own.
Why it matters to you
The benchmark proves the method behind our private work. If we can forecast public events well, and show our misses too, you can trust the forecasts we build for you.
Part 2 · Our results
Our record so far
Every round at a glance: how many questions, how many we got right, how good our probabilities were and how hard the set was. Open any round below for the full detail.
Round 1Marked
310 questions, 4 days ahead
- Scored
- 164
- Brier
- Not reported
Not measured
Round 2Marked
350 questions, 24 days ahead
- Scored
- 335
- Brier
- 0.165
50% chance a guess is right
Round 3Provisional
300 questions, 11 days ahead
- Scored
- 237
- Brier
- 0.198
42% chance a guess is right
To date
Rounds 1 to 3
- Questions
- 960
- Scored
- 736
- Correct
- 553
Totals to date cover Rounds 1 to 3. Round 1 is counted on its named-source basis (123 of 164). Round 3 is provisional. Hit-rate: share of scored calls we got right. Brier score: how good our probabilities were (lower is better; 0 is perfect; saying 50% on everything scores 0.25). Difficulty: the chance that a random guess is right (lower is harder).
Part 2 · Round by round
The questions, the answers and the difficulty
Each round shows what kind of questions we asked, how accurate we were, and how hard the set was. Round 3 is open below; tap any round to open it.
Round 1Locked 27 Jul, resolved 31 Jul 2026 · 4 days ahead75.0%123 of 164 against a named sourceMarked
The questions
310 predictions across nine areas, by share of the round.
20.3%16.5%12.3%11.9%11.6%9.7%6.8%5.8%5.2%
- Markets and macro: 63
- Culture, sport and corporate: 51
- Geopolitics: 38
- Space, tech and semiconductors: 37
- AI: 36
- Cyber: 30
- Energy and commodities: 21
- Climate: 18
- Health: 16
Difficulty
EasierHarder
Not measured. Round 1 questions were not tagged by type, so we cannot say how hard the set was. Rounds 2 onwards are.
How we did
Correct calls by area, counting every resolved call (269 of 310). Our headline rate, 75%, counts only the 164 calls checked against a named source.
Markets and macro was the one weak area, and the largest block of the round.
How sure we were, and how often we were right
Calls grouped by the confidence we stated before the outcome.
If anything we were slightly under-confident. The weakness was direction, not confidence.
Why the 41 misses happened
Every miss, sorted by cause.
What changed: in later rounds, market and economic calls must state a calm or recovery scenario, with its own probability, before they are locked.
Round 2Locked 7 Aug, resolved 30 and 31 Aug 2026 · 24 days ahead80.6%335 scored · Brier 0.165Marked
The questions
350 predictions: 150 foresight questions across society, technology, the economy, the environment and politics, plus 200 hard-number calls. Share of scored questions by type, from most to least predictable.
5.1%37.9%40.9%14.9%
- Scheduled events: 5.1%The date or outcome is largely set in advance.
- Predictable from history: 37.9%Past patterns give a strong guide.
- Noisy measures: 40.9%Numbers that move about, such as a monthly price or index.
- Head-to-head contests: 1.2%Which of several named rivals comes out on top.
- Shocks and surprises: 14.9%Rare or sudden events.
Difficulty
EasierHarder
60%50%40%30%
A random guess would be right 50% of the time. Most questions were predictable from history or were noisy measures; almost none were head-to-head contests.
How we did
Scored calls, split by how strongly we leaned one way.
What changed: from Round 3 we no longer make predictions at even odds. Every prediction must lean one way.
Hits and misses, as written
Six Round 2 predictions, locked on 7 August and due on 31 August 2026.
TSMC begins volume commercial shipping of 2nm (N2) chips.
- Our call
- 55% yes
- Outcome
- Yes
The 3GPP freezes Release 21, the full 6G specifications.
- Our call
- 5% yes
- Outcome
- No
A US commercial fusion facility delivers net electricity to the grid for 24 continuous hours.
- Our call
- 2% yes
- Outcome
- No
China's GDP growth for the first half of 2026 meets or exceeds 4.8% year on year.
- Our call
- 65% yes
- Outcome
- No
US utility-scale solar and wind exceed 22% of monthly generation in June or July 2026.
- Our call
- 55% yes
- Outcome
- No
Round 3Locked 19 Sep, resolved 30 Sep 2026 · 11 days ahead67.5%237 scored · Brier 0.198Provisional
The questions
300 predictions: 75 economic, 75 political, 75 technology, 38 social and 37 environmental. Share of scored questions by type, from most to least predictable.
Round 2
5.1%37.9%40.9%14.9%
Round 3
12.2%30.0%34.6%19.0%
- Scheduled events: 5.1% to 4.2%The date or outcome is largely set in advance.
- Predictable from history: 37.9% to 12.2%Past patterns give a strong guide.
- Noisy measures: 40.9% to 30.0%Numbers that move about, such as a monthly price or index.
- Head-to-head contests: 1.2% to 34.6%Which of several named rivals comes out on top.
- Shocks and surprises: 14.9% to 19.0%Rare or sudden events.
Difficulty
EasierHarder
60%50%40%30%
A random guess would be right 42% of the time, down from 50% in Round 2. A forecaster with our Round 2 skill would have scored 74.7% here: the set was 5.5 points harder, about 28% more expected errors.
How we did
Hit-rate on scored calls, by area. Provisional.
Economic was our weakest area, mainly short-term market calls during September's sell-off.
Where the skill shows
Provisional.
Why Round 3 scored lower
Our hit-rate fell 13 points, from 80.6% to 67.5%. Round 3 was a deliberately harder test, but we also got worse, and we are showing both.
- Predictable questions fell from 43% of the set to 16%.
- Head-to-head contests rose from 1% to 35%.
- 35% of questions had three or more possible answers. Round 2 had none.
- About 5 of the 13 points came from the harder questions. About 8 came from weaker forecasting on our part, mainly short-term market calls during September's sell-off.
- We asked very hard questions on purpose, to push and test our system properly. The full question set will be published in the public report.
Provisional: 15 calls are still awaiting mid-October data releases. 48 of 300 questions could not be fairly marked as written and were left out of the score (why). Method: each question is tagged with a predictability tier; Round 2 was tagged after the fact, without reference to its outcomes; the split of the fall uses a shift-share decomposition.
Rounds 4 to 6Locked by 20 Sep 2026 · resolve 31 Oct, 30 Nov and 31 Dec900predictions, locked and waitingOpen
These rounds test how far ahead our accuracy holds, from about six weeks to three months.
Round 4
Resolves 31 Oct 2026
About 6 weekspredicted aheadOpenRound 5
Resolves 30 Nov 2026
About 10 weekspredicted aheadOpenRound 6
Resolves 31 Dec 2026
About 3 monthspredicted aheadOpenThe questions
Each round holds 300 predictions, by area.
25.0%25.0%25.0%12.7%12.3%
- Economic: 75
- Political: 75
- Technology: 75
- Social: 38
- Environmental: 37
Difficulty
EasierHarder
Published with each round's results, on the same scale as Rounds 2 and 3.
Part 3 · Where we are going
From weeks ahead to years ahead
So far we have tested how accurately we predict up to about three weeks ahead. Rounds 4 to 6 test how far out that accuracy holds. Our goal is a system that can predict accurately years ahead, and we will show every step of the way, whether the score holds or falls.
How far ahead each round was predicted
From the date a round was locked to the date it resolves.
- Up to 3 weeksRounds 1 to 3, locked 4 to 24 days before they resolvedTested
- About 6 weeksRound 4, resolves 31 Oct 2026Open
- About 10 weeksRound 5, resolves 30 Nov 2026Open
- About 3 monthsRound 6, resolves 31 Dec 2026Open
- Years aheadOur longer-range predictions, the furthest due in 2051The goal
Longer range: 106 further predictions were locked by 19 July 2026, most taken from forecasts in our Codex of the Future. The first are due in October 2026 and the furthest in 2051.
The rules
How we score, and how we show the misses
The scoring rules were fixed before any result was known, and they apply to every round.
What counts as a hit
Every prediction carries a probability. 50% or above means we said it would happen; below 50% means we said it would not. It is a hit when the named source shows that side happened.
Saying no counts too
A correct "it will not happen" counts the same as a correct "it will happen". But easy no's can flatter a score. Counting every resolved call, Round 1 scored 269 of 310. We lead with the harder figure: 123 of 164 checked against a named source.
Held, never guessed
If the named source cannot settle a prediction, it is held and left out of the score. In Round 2, 15 of 350 were held. In Round 3, 15 are waiting for official data.
Questions we could not fairly mark
Round 3 exposed a gap: some questions were not checked against their sources before they were locked. 48 of 300 could not be fairly marked as written, so we left them out of the score rather than mark them either way. Every round is now checked automatically before it is locked, and the affected questions in Rounds 4 to 6 have been replaced before any of them resolve.
Misses stay on the record
Nothing is edited after a round is locked. A changed view is a new, dated prediction. In Round 1 our weakest area was markets and the economy, at 42 of 63, and we changed how we write those predictions.
Harder rounds, on purpose
We vary how hard each round is, and we show it. A higher score on easier questions proves less than a lower score on hard ones, so every round is published with its difficulty.
Not investment advice
The benchmark tests our forecasting. It is not a recommendation to buy or sell anything.
Answers
Common questions
Why did Round 3 score lower than Round 2?
Our hit-rate fell 13 points, from 80.6% to 67.5%. Round 3 was a deliberately harder test: predictable questions fell from 43% of the set to 16%, and the chance of guessing right fell from 50% to 42%. About 5 of the 13 points came from the harder questions and about 8 from weaker forecasting, mainly short-term market calls during September's sell-off. Round 3 is provisional.
Why does the 311 Institute publish its predictions?
Because the best way to help our clients is to predict the future accurately, and the only honest way to show that is in public. Our proprietary system predicts global social, technology, economic, environmental and political events. We publish those predictions before they happen and mark them against named public sources, so anyone can check our record.
Do you publish all your predictions?
No. We publish predictions about public world events so our accuracy can be tested. The predictions and foresight we create for clients stay private: they are our intellectual property and theirs, and are never published where anyone could copy them.
What is the 311 Prediction Benchmark?
It is the public test of 311 Institute's forecasting. We write down dated predictions with a probability, lock them before the outcome can be known, then mark every one against a named public source and publish the hits and the misses together.
How do you stop predictions being changed after the event?
Each round is sealed before its resolution date and nothing in it is edited afterwards. If our view changes, that is a new, dated prediction. It never replaces the old one.
What counts as a correct prediction?
Every prediction carries a probability. 50% or above means we said it would happen; below 50% means we said it would not. It is a hit when the named source shows that side happened. Saying something will not happen counts the same as saying it will.
Why is Round 3 marked provisional?
Round 3 is provisional because 15 of its calls are still waiting for official data releases due in mid-October. Its figures will be updated once those calls are marked.
What happens when a prediction cannot be checked?
It is held, not guessed. In Round 2, 15 of 350 predictions were held because the named source could not settle them. In Round 3, 48 of 300 questions could not be fairly marked as written and were left out. Held and voided questions are never counted either way.
How often are new results published?
A new round of 300 predictions comes due at the end of each month from September to December 2026. Results are added here once each round is marked and checked.
Put a tested forecaster in front of your board.
Tell us what you are facing and we will show you what our forecasting says about it, and how sure we are.