Forecasting · Prediction Markets · Geopolitics · Methods BriefInstruments read 21 Sep 2026
Four instruments, one shelf, and not one of them answers the question
Nobody Can Tell You the Odds
Ask how likely a world war is and you will be pointed at a handful of numbers: a crowd forecast, a prediction market, the Doomsday Clock, an index of geopolitical risk. This page tried to read all of them on one day. The most-quoted of them refuses to open without an account. The most heavily traded world-war market on the open platform is a user-written question indexed to the release of a video game. The Clock is a board’s argument about leadership, published once a year. The risk index counts newspaper articles. And the two instruments that can be read cleanly measure only what has already happened — one of which says, flatly, that global conflict held steady.
What Happened
What each instrument returned when asked Every request below was made on 21 September 2026. Response codes are facts about access on that day, not about the instruments’ contents.
- 403The response from the most-quoted world-war forecast, on the page and on both of its API endpoints
- 590Bettors on the most-traded world-war market found — a question about a video game release
- 85Seconds to midnight on the Doomsday Clock, set once a year by a board
- 204,605Conflict events ACLED counted in a year — slightly fewer than the year before
| Standing | What | On whose record |
|---|---|---|
| Confirmed | The most-quoted crowd forecast cannot be read. Metaculus’s long-horizon world-war question returned HTTP 403, and both API endpoints answered: “Permission Error: The API is only available to authenticated users.” No Metaculus figure appears anywhere on this page as a result. | This session’s own requests |
| Confirmed | The open market is open, and thin. Manifold’s public API answered without authentication. The most heavily traded world-war market found was “Will WW3 happen before GTA6?” at 5.4% with 590 bettors; the next, “Will WW3 start before gta6?”, at 4.0% with 30. | Manifold Markets |
| Confirmed | The Doomsday Clock stands at 85 seconds, set 27 January 2026 and described by its own board as the closest it has ever been. The sequence at source is 100 seconds in 2020, 90 in 2023, 89 in 2025, 85 in 2026. | Bulletin of the Atomic Scientists |
| Confirmed | The Clock is not a forecast and says so in its own words. Its 2026 statement gives as its reasons that major countries have become “increasingly aggressive, adversarial, and nationalistic” and a “failure of leadership”, naming New START’s expiry, the Golden Dome missile-defence plan and possible resumption of US nuclear testing. | Bulletin of the Atomic Scientists |
| Confirmed | Global conflict did not reach a record. ACLED counted 204,605 conflict events in the twelve months to 28 November 2025 against 208,219 in the twelve before, and states that conflict “remained at a steady level”, with deaths conservatively above 240,000. | ACLED |
| Confirmed | One theatre’s casualties genuinely rose. UN monitors verified 2,222 civilians killed and 13,058 injured in Ukraine from January to August 2026, 55% above the same months of 2025. Both things are true at once, of different scopes. | UN Human Rights Monitoring Mission in Ukraine |
| Confirmed | The geopolitical risk index is a spreadsheet. Its page states the monthly series is updated at the start of each month and the daily series every Monday, last updated 14 September 2026. The export file downloads — 2,714,112 bytes of legacy spreadsheet — and would not open with the tooling here, so no value from it appears on this page. | Caldara and Iacoviello’s own site |
| Unconfirmed | A long-horizon estimate of about one in three for a new great-power war within roughly 30 years, and 0.05% to 2% for an existential catastrophe from war this century. These are one organisation’s stated judgements, published as such. | 80,000 Hours |
| Unconfirmed | A figure near 45% over about 77 years, or 22% cumulative by 2050, circulates attributed to that same organisation. The page read here says one in three over roughly 30 years. This record carries the discrepancy rather than reconciling it. | unattributed circulation, against the page itself |
| Guesswork | That treating these as one dial labelled “risk of world war” is the error — because they are a crowd forecast, a play-money market, a symbolic annual statement, a measure of press coverage and two backward-looking counts. This page’s reading, and marked as one. | This page’s inference |
| Not public | What the Metaculus community currently forecasts, and whether it moved over the past year. A figure exists; this page cannot see it, and will not borrow one from a summary. | — |
The standing column separates two things that look alike. The response codes are the hardest evidence on this page — a 403 is not a matter of interpretation, and anyone can reproduce it. So are the published series: ACLED’s counts and the UN’s tallies state their own scope and window. What is softer is every number that is somebody’s judgement, and this page is careful to say whose.
- 403returned by the page and both API endpoints of the most-quoted forecast
- 5.4%the price on the most-traded world-war market found, which is about a video game
- 2.7 MBof legacy spreadsheet that is how the geopolitical risk index publishes its numbers
Timeline
Six years of a clock, one index, and one afternoon of requests Thin on purpose. A subject made of instruments has little chronology, and saying so is part of the finding.
- 2020The Doomsday Clock is set to 100 seconds to midnight.
- 2023Set to 90 seconds.
- 2025Set to 89 seconds — a one-second move, which is a message rather than a measurement.
- 11 Dec 2025ACLED publishes its Conflict Index for the twelve months to 28 November: 204,605 events against 208,219, and the judgement that conflict “remained at a steady level”.
- 27 Jan 2026The Clock is set to 85 seconds, the closest ever, for a “failure of leadership” and the expiry of New START.
- 14 Sep 2026The geopolitical risk index’s daily series is last updated. Its numbers ship as spreadsheets.
- 16 Sep 2026UN monitors publish Ukraine’s August civilian-casualty figures, showing the year already past all of 2025.
- 21 Sep 2026This record. Metaculus refuses three requests; Manifold answers one; the risk index’s file downloads and will not open. Every instrument on this page was asked on this day.
A one-second move is not a one-second measurement
From 90 seconds in 2023 to 89 in 2025 to 85 in 2026 looks like a series, and reading it as one is the mistake. The Clock is set annually by a board publishing an argument, and its stated reasons in 2026 are about treaties, weapons programmes and leadership — not a calculation. It is the most legible instrument here and the least numerical, and those two facts are connected.
What has no date at all
The geopolitical risk figures, because the file would not open. Any movement in the crowd forecast over the past year, because the forecast could not be read. And the origin of a long-horizon percentage that circulates attributed to an organisation whose own page states something different. None of the three is missing a date; all three are missing a reading.
The Argument
A year of new wars need not move a great-power question Argued from the instruments’ own wording and from two published series. The classification is marked as this page’s reading.
The complaint people reach for is that forecasters are complacent: the world plainly got worse, so why did the numbers not move? Two published series answer it without any appeal to psychology. ACLED counted slightly fewer conflict events in the year to November 2025 than in the year before — 204,605 against 208,219 — and said conflict held steady. UN monitors recorded Ukrainian civilian casualties up 55% over eight months. Both are correct. They are measuring different things at different scopes, and neither is measuring what a world-war question asks about.
A world-war question resolves on direct combat between great powers. Regional wars, however large and however many, do not satisfy it. So a year can add conflicts, casualties and newspaper coverage while leaving that specific question untouched — not because anyone is being blasé but because the question is about a particular thing and those are not it. The failure is one of reading, not of forecasting.
Which leaves the harder problem, and it is about access rather than accuracy. The instrument best suited to the question is the one a reader cannot open. A crowd forecast with an explicit resolution rule is exactly the right tool; it answered three requests with a permission error. What remains readable is a play-money market whose busiest world-war question is pegged to a video game, a board’s annual statement, and a spreadsheet of newspaper counts. A public debate conducted on those terms is not short of numbers. It is short of one.
- 204,605 / 208,219global conflict events, latest year against the year before — slightly down
- +55%Ukrainian civilian casualties over the same kind of period — sharply up
- 1 of 4instruments that is actually a forecast, and it is the one behind a login
What argues against this page
Access is not the same as absence. Metaculus has a figure, a community and a resolution rule, and plenty of people can see all three; what this page establishes is that a reader without an account cannot, on the day tested. Manifold’s prices, thin as they are, are consistently in the low single digits, which is itself a signal worth having. And a symbolic instrument is not a useless one — the Clock’s reasons are specific, checkable and about real treaties.
A different question, answered elsewhere
Whether the wars now running already amount to a world war is a question about definitions and about what those wars share. It is not this page’s question and is deliberately not re-argued here. This page asks only what the instruments are and what each of them measures — and finds that the answer differs for every one of them.
What Others Add
Four instruments, side by side What each one is, what it actually measures, and whether a reader without an account can see it.
| Instrument | What it measures | Readable without an account? |
|---|---|---|
| Crowd forecast | A probability, with an explicit resolution rule | No — 403, authentication required |
| Play-money market | Prices set by participants on questions they wrote themselves | Yes — open API, 30 to 590 bettors per market |
| Doomsday Clock | A board’s published argument, set once a year | Yes — statement and full timeline both open |
| Geopolitical risk index | How often newspapers discuss geopolitical risk | Partly — the page opens, the numbers ship as a 2.7 MB legacy spreadsheet |
| Conflict event count | Violence that has already happened, worldwide | Yes — and it says the latest year held steady |
| Civilian casualty monitoring | Verified civilian harm in one country, monthly | Yes — and it states its own verification gap |
The two backward-looking instruments are the best-behaved on this page, and it is worth saying why: each states the limits of its own knowledge. ACLED gives its window explicitly — twelve months to 28 November 2025 — so nobody can read it as a statement about 2026. The UN mission verified four civilian deaths in occupied Ukraine in August while noting Russian authorities described at least 85, and says plainly that it cannot verify most allegations. An instrument that publishes its own blind spot is easier to use than one that publishes only a number.
The market
The busiest world-war question is about a video game
- “Will WW3 happen before GTA6?” priced at 5.4% with 590 bettors — the most-traded of those found.
- A second, nearly identical question sat at 4.0% with 30 bettors — the same subject, a twentieth of the participation.
- The prices are informative and consistent; what they are not is a measurement of the world, and the bettor counts are given so a reader can judge.
The index
It counts articles, not dangers
- The geopolitical risk index measures how often newspapers discuss geopolitical risk, so it rises when coverage rises.
- In a year with new wars, coverage rises for reasons that have nothing to do with great-power combat.
- Its own page says the monthly series updates at the start of each month and the daily series every Monday; no value from it appears here, because the file would not open.
The long view
One organisation’s stated estimate, and a figure that differs from it
- Its page states about one in three for a new great-power war within roughly 30 years.
- And 0.05% to 2% for an existential catastrophe caused directly by war this century.
- A figure near 45% over about 77 years circulates attributed to the same organisation. This page carries the discrepancy and does not reconcile it.
Access is a property of an instrument, not a footnote about it
A forecast nobody outside can read cannot do the public work a forecast is for. That is not a complaint about paywalls in general — a platform may require accounts for good reasons — but it changes what the rest of us are arguing with. On the day tested, the public debate about world-war risk had available to it a joke market, an annual editorial and a spreadsheet, and the one purpose-built instrument answered with a permission error.
Conclusion
How to read a number about a war that has not happened Five things this record supports, in the order a reader is likely to need them.
- Ask what the instrument is before you read its number. A crowd forecast, a play-money market, an annual board statement and a count of newspaper mentions are four different kinds of object. Only the first is a probability of the thing being asked about, and it is the one behind a login.
- Check the participation before you trust a market price. Thirty bettors and 590 bettors are different objects wearing the same percentage sign. On the platform read here, the busiest world-war question was pegged to a video game release — which is informative about the crowd, not about the world.
- Do not read the Doomsday Clock as a scale. It is a board’s published argument set once a year, and its 2026 reasons are treaties, weapons programmes and a stated failure of leadership. Reading 89 to 85 as a four-unit rise in danger is reading an editorial as a thermometer.
- Keep scopes apart. Ukraine’s verified civilian casualties rose 55% in eight months while ACLED’s global conflict-event count fell slightly and the index called the year steady. Collapsing those into one sentence about the world getting worse discards the only two things here that were actually counted.
- And expect a year of large wars to leave a great-power question unmoved. Those questions resolve on direct combat between great powers; regional wars do not satisfy them however many there are. When a forecast does not move after a dramatic year, the first hypothesis should be that it is answering a narrower question than the reader assumed.
What this page could not see, stated rather than filled
No crowd-forecast figure appears here, current or historical, and no geopolitical risk value. Both exist; neither could be read on the day. The tempting move was to borrow a number from a secondary summary and present it as the instrument’s reading, which is precisely the failure this page is about. Absence of access is not absence of a number, and the two are kept distinct throughout.
And it can be repeated
Every reading here is a request anyone can make again: three URLs that returned a permission error, one open API query with its prices and bettor counts, an annual statement and its timeline, a spreadsheet that downloads, and two published indices with their windows stated. If the access picture changes, the finding changes with it, and that is the point of writing the response codes down.