World War 3 and a
theory of relative decline.
This is the most disciplined page on the site, and it has to be. A domain like this one earns trust only by refusing to do the thing the domain implies. So: no prophecy, no inevitability, no world war. Conditionals with named triggers, probability bands with resolution dates, and public scoring afterwards.
Updated 22 October 2026
This page does not predict a world war. It argues that relative decline raises the risk of conflict through “window logic”: a power whose position is deteriorating on a known schedule may act while it still can. Five observable conditions are scored against eight flashpoints, and every probability band is published with a resolution date.
Key facts
- The Uppsala Conflict Data Program (UCDP) recorded 65 state-based conflicts in 2025, the most since its records began in 1946.
- Eight of those 2025 conflicts were interstate, up from two in 2023 and the highest count since 1946.
- UCDP counted about 244,600 deaths from organised violence in 2025, the second-deadliest year since the 1994 Rwanda genocide.
- UCDP recorded about 76,500 deaths from one-sided violence in 2025, an increase of more than 400% and the highest since 1994.
- UCDP counted 13 conflicts with at least 1,000 battle-related deaths in 2025, the most since 1992.
- The probability bands on this site are subjective five-year estimates of a state-on-state clash above 1,000 battle deaths, labelled Low confidence, and each is scored at its resolution date.
The mechanism: window logic
The popular model of decline-driven conflict is desperation — a failing state lashing out. That model is mostly wrong, and it misdirects attention toward the weakest powers when the risk sits with the strong ones.
The better model is window logic. A power whose relative position is deteriorating on a known schedule faces a choice between acting while it still can and accepting a permanently worse settlement later. It is not desperate. It is doing arithmetic. And the arithmetic gets worse every year it waits, which converts patience from a virtue into a cost.
Demographic decline is uniquely dangerous inside this model because it is the most legible form of decline there is. Fiscal trouble can be inflated away, argued about, or deferred. Productivity is contested and revised. But a leadership can read its own birth cohorts and know, to the year, when the manpower, the tax base and the strategic option all close. There is no ambiguity and no policy that reverses it inside a decade.
The five conditions
Each is independently observable, and each flashpoint on the board below is scored against all five. The claim is not that five-of-five means war. It is that the historical base rate rises sharply with the count, and that a case meeting all five deserves a completely different level of attention from one meeting two.
- A closing window, visible to the leadership that holds it. Not merely decline — decline with a date on it. China’s military-demographic window across roughly 2027–2035; Russia’s manpower window, already half shut.
- A dissatisfied riser. This is the Organski variable and it is the discriminator that most commentary drops. The United States overtook Britain without war because it was a satisfied riser that accepted the existing order. Power transition alone predicts nothing; dissatisfaction is what makes transition dangerous.
- A contested fixed resource. Minerals, water, sea lanes, seabed, orbital slots. Fixed supply is what makes bargaining fail, because there is no version of the deal where both sides get more.
- An eroded guarantee. The defender’s commitment is doubted by both the challenger and the protected party. This is the classic deterrence-failure condition, and it is doubted symmetrically — which is why reassurance and deterrence pull against each other.
- Domestic legitimacy pressure. Diversionary incentives, which rise precisely as the cohesion engine fails. In the index, the cohesion engine is the fastest-moving of the five — it can crack in a decade — which makes it the one most likely to change a flashpoint’s score between quarterly reviews.
The flashpoint board
How to read the bands
The probability bands are subjective five-year estimates of a state-on-state armed clash exceeding 1,000 battle deaths. They are the author’s judgement, not a model output, and they are labelled Low confidence throughout for that reason. Each will be scored at its resolution date and the record published — including the misses, prominently.
Three points on interpretation that matter more than the numbers:
- High probability does not mean high consequence. The South China Sea carries a higher band than Taiwan and a far lower escalation ceiling. Hormuz carries the highest band on the board and is the case the world has already absorbed repeatedly. Read the band and the character together or you will read both wrong.
- The Sahel entry is not a forecast. It is already the largest fatality source in the dataset. It sits on the board because a site about great-power decline that only counts great-power casualties is measuring the wrong thing.
- Greenland is on the board at under 5% because acquisition contests run through licences, export credit and basing rights long before they run through force. Low probability of shooting is not low significance.
The critiques, carried openly
Power-transition theory and its popular form, Graham Allison’s Thucydides Trap, are the intellectual engine of this page, and the objections to them are strong enough that presenting the framework without them would be dishonest.
- Determinism. The framework is routinely read as predicting an inevitable clash. It does not, and this site does not. It identifies conditions under which war becomes significantly more probable. A hazard rate is not a destiny.
- Measurement. Allison’s sixteen-case file has been criticised for non-uniform measurement and selection effects, and the veneer of scientific rigour it carries is not matched by replicable standards. Worse, “satisfaction” — the variable that does most of the work — is genuinely hard to operationalise. Power is measurable in GDP and budgets. Contentment with the international order is not.
- Counterexamples. Several transitions have passed without war, most notably the United States overtaking Britain. Any honest use of the framework has to explain the peaceful cases, not just the violent ones.
- Self-fulfilment. The most serious objection. If policymakers believe war is fated, they take precautionary actions that are indistinguishable from preparation, and the other side responds in kind. A site that prices conflict has to be conscious that pricing is itself an input.
Which is why the framing on this page is narrow and deliberate. The claim is not that a world war is coming. It is that the conditions that have historically preceded great-power war are present in more places, simultaneously, than at any point since 1945 — and the observable that makes that non-hysterical is the data at the top of this page: 65 state-based conflicts, eight of them interstate, 244,600 dead in a single year and rising.
Interdependence, nuclear weapons and demographic exhaustion all cut the other way. The powers most often named as aggressors are precisely the ones running out of young men — a peaking power with a collapsing birth rate has fewer soldiers each year, not more, and mass-casualty ground war is the one instrument its demography forecloses. Nuclear deterrence between all the major dyads has held for eighty years including through several crises worse than anything currently live. And the 2026 Hormuz episode demonstrated the interdependence argument as much as the fragility one: the conflict transmitted into bond markets within weeks, which is a very expensive disincentive.
The UCDP numbers also need care. The record conflict count is driven substantially by fragmentation of non-state actors and by improved data coverage, not solely by more great-power aggression, and the fatality spike is overwhelmingly Sudan rather than any great-power theatre.
The forecast ledger
This is the feature that will decide whether the site is worth reading in three years, and it is the one it should be judged on. Every band above resolves. Every resolution gets published — hit or miss — with the original text unedited and the score attached. Where base rates allow, resolutions are Brier-scored.
A site that grades its own predictions in public can be trusted with a domain like this one. A site that does not, cannot, and the domain becomes the argument instead of the evidence.
Frequently asked questions
How does national decline lead to war?
Through window logic rather than desperation. A power whose relative position deteriorates on a known schedule chooses between acting while it still can and accepting a permanently worse settlement later. Demographic decline is the most dangerous kind because it is the most legible: a leadership can read its own birth cohorts and date the closing of the option. See the China and Russia chapters.
What are the conditions that precede great-power war?
Five, each independently observable: a closing window visible to the leadership holding it; a dissatisfied rising power; a contested fixed resource; an eroded security guarantee doubted by both sides; and domestic legitimacy pressure creating diversionary incentives. The theory page sets out the literature behind them.
How many armed conflicts are there in the world?
UCDP recorded 65 state-based conflicts in 2025, a record since records began in 1946, including eight interstate conflicts, double the previous year and the highest since 1946. Fatalities reached 244,600, up from roughly 187,000 in 2024. The steelman above explains why the headline count needs care.
Does this site predict that World War 3 will happen?
No. It prices conditions and publishes conditional probability bands with resolution dates, and it carries the main critiques of power-transition theory openly. How the index is scored is on the method page; who outlasts whom is on the outlast page.
Sources
- Uppsala University, UCDP press release: record number of conflicts between states (9 June 2026)
- Uppsala Conflict Data Program
Continue
The five engines and the literature ·
How the index is built ·
Editorial standards and corrections ·
Europe · United States