Collapse Warning scenario laboratory
Explore how regional essential systems respond to interacting constraints from 2025 to 2100. Choose a response portfolio and world conditions, then compare the resulting conditional futures with documented Business as usual.
This is a scenario laboratory, not a forecast. Frequencies describe repeated runs of this model under selected assumptions; they are not observed probabilities of humanity’s future.
Audit status: diagnostic, review-ready implementation. The simulation core has not been certified by independent scientific review, and no public claim should rest on it until it has been.
The question this model answers
What it asks
Within this model, under a chosen set of structural assumptions, parameter ranges, response settings, shock rules and random seed: how often do nine world macroregions reach an operational definition of severe simplification between 2025 and 2100, and which physical and social systems give way first?
A region is counted as affected only when a criterion below holds for ten consecutive years and material welfare is declining at the same time. Once a region has confirmed such an interval it stays counted for the rest of the horizon.
The headline numbers are shares of the world’s 2025 population: the 2025 population of every affected region, divided by the 2025 world total. A region enters that share whole, in the year its interval is confirmed — so a result of 30% means that regions holding 30% of the world’s people in 2025 have reached the definition, not that 30% of individuals each crossed a threshold. Populations are held at their 2025 values on purpose: otherwise a region losing people would make the share of the affected fall.
What it does not answer
- It does not estimate a real-world probability. Every number on this page is a frequency across repeated runs of one uncertain model.
- It is not about human extinction. Severe regional simplification and the end of the species are different questions, and only the first is modelled here.
- It does not predict policy adoption. Ambition settings describe additional installed coverage sought and cap annual mobilization; the response ledger decides what can be built, maintained and retained.
- It is not World3 and does not reproduce it. The trajectories here come from an independently specified regional model.
The four severe-simplification criteria
Each criterion must hold for ten consecutive years, alongside declining material welfare, before a region is counted as affected.
- Energy services. Final energy services per person at least 30% below the region’s own 2020–2025 baseline.
- Essential services. Loss of nationally scaled reliable distribution, represented by utility reliability below 65% of the region’s baseline.
- Excess mortality. Non-combat mortality at least 20% above the demographic counterfactual anchored to the baseline. The counterfactual is age-adjusted: a literal fixed crude-death-rate comparison would classify ordinary population ageing as collapse late in the century, so the model does not use one.
- GDP per person. Real GDP per person at least 25% below the region’s own baseline.
How the model works
Nine non-overlapping macroregions each hold physical and social stocks: energy capacity, industrial capital, infrastructure, soil and water, food stores, health capacity, institutional capacity, legitimacy, skilled labour and ecological condition. Three traded essentials — food, usable energy and industrial components — are cleared across a trade network each year.
Every region has ten response accounts: energy, fiscal capacity, industry, materials, infrastructure, skilled labour, institutions, political bandwidth, international cooperation and ecological headroom. Ongoing crises commit part of each account before anything new can be attempted. The seven interventions you choose all draw on the same remaining balances, and a unit cannot be spent twice.
Each run samples uncertain parameters, generates conditional shocks with a duration, a severity and a regional reach, integrates the stocks annually to 2100, and records which criteria were met. Repeating that many times with different draws from the same seed gives the spread of outcomes reported on this page.
Every run in this page happens inside your browser, in a background worker. Nothing is sent to a server.
Documented starting points
Each preset is defined by the model, not by this page. Inspect one before loading it, then change anything you like: the builder below always shows the settings that will actually be run. Business as usual is grounded in revealed implementation and current-policy behaviour, without assuming that unfulfilled targets or pledges will suddenly be delivered.
Build a scenario
1. Response portfolio
Seven sustained interventions. Each card carries two different controls. Ambition is the additional system coverage sought and the ceiling on normalized annual mobilization to close the remaining gap. Priority is a proportional scarcity weight for maintenance and new build, not a first-to-last funding order. Raising ambition does not create capacity, and a 100% goal does not mean 100% will be installed: the response ledger decides what can be built, maintained and retained, and the result appears on each card after a run.
2. Cross-cutting choices
Three settings that change the conditions every intervention operates in.
3. External conditions
Conditions imposed on the world rather than actions taken in it.
4. Model uncertainty
These settings change the model itself, not the world it describes. The three structural variants are competing families of equations. They are not confidence levels, and they are not optimism and pessimism.
Ready. Every run happens in your browser.
Conditional results
This is a scenario laboratory, not a forecast. Frequencies describe repeated runs of this model under selected assumptions; they are not observed probabilities of humanity’s future.
No scenario has been run yet. Choose your settings above and run the model to fill this section.
Settings have changed since these results were produced. Everything below still describes the earlier run; run the scenario again to update it.
Four headline outcomes
Share of world population in affected regions
System trajectories, 2000–2100
Every panel shares one time axis with a marked boundary at 2025. To the left of it are sourced observations, wherever the model has them. To the right are the model’s median and uncertainty band. Indices that exist only inside the model start at 2025 and are never back-cast, which is why their left half is blank. Each indicator keeps its own axis and unit; the year cursor is shared across every panel.
Material and production systems
Critical material classes
The aggregate material curve can hide the constraint that actually binds. These four stocks affect food, maintenance, industrial components and clean-energy construction through different channels.
Land competition
A finite regional land stock is shared between direct food crops, livestock feed and grazing, bioenergy, settlements and ecological habitat. Diet and aviation therefore compete with food and ecosystems; land changes use gradually rather than instantly.
People and response capacity
Essential manufacturing continuity
The share of the world’s 2025 industrial output held by regions that can still sustain each capability. This is continuity of the ability to manufacture, not consumption and not production volume.
Carbon and climate
System state at the focus year
Regions
The map, the table below it and the regional breakdown all read the same records for the focus year and region set above, so they always agree. Countries the model does not place in a macroregion stay neutral grey and are listed under the table rather than attached to a neighbour.
Which criterion is responsible
Response ledger
For the focus region and year: sustainable capacity, finite reserves, conserved transfers, required obligations, unmet work and the portfolio allocation. Both available capacity and required work reconcile exactly; shortages persist as arrears instead of disappearing between years.
This scenario against Business as usual
Shock exposure
How often the model generated each kind of event under these assumptions, and which regions were inside its reach. These are exposure diagnostics, not observed event probabilities.
Raw response and reproducible state
Limitations and provenance
- Nine macroregions are a severe simplification. Within any one of them, conditions differ enormously between places, and a regional average can hide both catastrophe and continuity.
- An annual timestep cannot represent fast cascades: a grid collapse, a bank run or a panic in food markets happens far below this resolution.
- Adaptation, politics and conflict are represented by aggregate parameters. Real institutions do things no parameter in this model can express.
- Shock rules are calibrated on a short and unrepresentative historical record, and the correlation between shocks is deeply uncertain.
- The frequencies here are conditional on this model, this data, these distributions and the settings you chose. Two structural variants can disagree more than the whole spread of runs within either of them.