H Heuristics · Digital Report № 2026-24 · October 2026
How a global polycrisis could spiral into civilisational collapse lasting centuries — and what keeps repair ahead of stress
Not a single catastrophe but a race: crises add stress, civilisation's capacity repairs it, and stress erodes the capacity that does the repairing. When the race is lost, history suggests the regression that follows is measured in centuries.
This report defines and examines the Polycrisis Apocalypse, a term coined by H Heuristics for a scenario in which multiple interconnected global crises intensify and reinforce one another beyond humanity's capacity to manage them, triggering cascading systemic failures that end in the collapse of industrial civilisation and a prolonged period of societal regression lasting centuries or millennia. Unlike a conventional apocalypse, it has no single cause or moment. It is a process, and the report argues that its decisive variable is the balance between two rates: the rate at which interacting crises add stress, and the rate at which a civilisation's institutions, infrastructure and finance repair it.
The historical record shows that such regressions are common and slow. In the Mycenaean world, writing disappeared with the palaces around 1200 BCE and did not return for roughly four and a half centuries; Roman-era lead pollution recorded in Greenland ice fell with the Antonine Plague of 165 CE and did not regain its level for more than 500 years; Egypt's First Intermediate Period lasted about 125 years. Survival analysis of premodern states finds that their risk of termination rises over the first two centuries of their existence, and one survey of past civilisations puts their average lifespan at about 336 years. Collapse, in this record, is a loss of complexity rather than of people, and recovery is measured in generations.
An interactive model makes the mechanism explicit. Crises arrive at random and add stress; capacity repairs stress but saturates when stress is high; stress erodes capacity; and amplification between crises depends on how interconnected the system is. The model has a sharp threshold. At its default settings, 91 per cent of 300 simulated runs collapse within 1,000 years and the median dark age lasts 471 years, but raising repair capacity by a sixth cuts that share to 27 per cent, loosening coupling cuts it to 10 per cent, and a combination of levers removes it entirely. Faster rebuilding, through preserved knowledge and institutions, does not prevent collapse but reduces the share of time spent collapsed from 51 to 19 per cent. The model is illustrative, not a forecast.
Today's ledger shows stress rising on most of the ten crisis domains in the definition — seven of nine planetary boundaries breached in 2025 against three in 2009, 65 state-based armed conflicts in 2025, 673 million people hungry in 2024 — while several measures of repair capacity weaken: official development assistance fell 23.1 per cent in real terms in 2025, the largest drop on record, and global human development gained 2.75 thousandths a year in 2019–2023 against 5.6 in the 2000s. Modern societies still recover fast, as the return of life expectancy after the pandemic shows. The report concludes that the Polycrisis Apocalypse is avoidable precisely because it is a race: the levers that keep repair ahead of stress — reducing shocks at source, loosening the couplings between systems, sustaining the capacity to repair, and preserving the means to rebuild — are known, and the model shows they compound.
A name for the worst way the polycrisis could end, and a framework for keeping it hypothetical.
Polycrisis Apocalypse (noun)
A scenario in which multiple interconnected global crises intensify and reinforce one another beyond humanity's capacity to manage them, triggering cascading systemic failures that ultimately result in the collapse of industrial civilization and a prolonged period of societal regression lasting centuries or millennia.
Term coined by Hunter Hughes, H Heuristics, 2026
The word apocalypse usually summons a single event: an impact, a war, a plague that arrives on a known day. The Polycrisis Apocalypse is different in kind. It has no single cause and no single date. It unfolds through the interaction of many crises — climate and ecological degradation, resource depletion, water scarcity, food insecurity, economic instability, geopolitical conflict, demographic change, infrastructure decay, technological disruption and pandemics — each of which amplifies the others until the institutions, governments and supply chains that hold a complex civilisation together are overwhelmed. Its signature is not the moment of failure but the length of what follows: a regression measured in centuries.
This dashboard treats the concept as a hypothesis to be examined rather than a prophecy to be believed. It asks three questions. Has anything like it happened before, and how long did the regression last? What mechanism would turn a polycrisis into a collapse rather than a bad decade? And where does the world stand today on the quantities that mechanism depends on? The answers come from the historical record, from an interactive model that readers can run for themselves, and from a ledger of current indicators for each of the ten crisis domains in the definition.
The argument in brief
A Polycrisis Apocalypse is a race that is lost. Crises add stress to a civilisation; its capacity — institutions, infrastructure, finance, cooperation — repairs that stress; and stress, left unrepaired, erodes the very capacity that does the repairing. Below a threshold, a society absorbs shocks and recovers within years. Beyond it, the loop runs the other way, and the past suggests that what is lost then takes centuries to rebuild. Today stress is rising on most fronts while several measures of repair are weakening. That is a warning, not a verdict: the race can be won, and the levers that win it are known.
The distinction between a bad century and an apocalypse matters for what should be done. A bad century calls for crisis management: absorb each shock, recover, move on. A Polycrisis Apocalypse calls for something harder — protecting the capacity to recover itself, because that capacity is what the spiral consumes. Much of the policy agenda of recent years, from adaptation finance to debt relief to pandemic preparedness, can be read as an attempt to do exactly that, and the evidence in this dashboard is that the attempt is now being underfunded at the moment it matters most.
The sections that follow define the concept precisely and distinguish it from its neighbours (section 1), review what past collapses lost and how long recovery took (2), model the spiral and its threshold (3), set out today's ledger of stress across the ten crisis domains (4), look for early warnings that repair is slowing (5), and translate the model's levers into off-ramps (6). The method section states every assumption the model makes; none of its numbers should be read as a forecast.
Three conditions separate a Polycrisis Apocalypse from a run of bad years, from a single-event catastrophe and from human extinction.
The first condition is interaction. The crises involved must reinforce one another rather than simply coincide: drought raising food prices, food prices fuelling unrest, unrest disrupting trade and energy, energy shocks deepening debt, debt cutting the budgets that would have paid for adaptation. This is the Cascade Institute's sense of a global polycrisis — crises that become causally entangled so that their combined harm exceeds the sum of their parts — and it is what distinguishes the concept from a run of independent misfortunes.
The second condition is overwhelm. Interacting crises are a feature of every century; what makes this scenario apocalyptic is that their combined load exceeds the capacity of institutions to manage them, so that failures cascade instead of being contained. The definition locates the tipping point not in any single crisis but in the relation between load and capacity. A world with far more stress than today could remain stable if its capacity grew faster; a world with less could collapse if its capacity eroded.
The third condition is duration. The outcome is not a recession, a war or a pandemic from which the world rebounds within a decade, but the collapse of industrial civilisation followed by regression lasting centuries or millennia. That is what distinguishes it from the bad decades of the twentieth century, and also from human extinction, which is a different and more final category of risk: in a Polycrisis Apocalypse people survive, but the complexity they had built — its knowledge, networks and institutions — does not, and must be rebuilt slowly.
The spiral: ten crisis domains around a shared capacity to manage them
When the load arriving at the centre exceeds what it can repair, capacity shrinks, repair falls further and reinforcement around the ring strengthens. Section 3 turns this loop into a model.
What it is
What it is not
Framed this way, the concept connects two literatures that rarely meet. The first is the study of past collapses, from Joseph Tainter's argument that societies falter when the returns on their complexity decline, to recent survival analyses of premodern states. The second is the modern study of systemic and catastrophic risk, which maps how shocks travel between climate, food, finance, conflict and health. The Polycrisis Apocalypse sits where they overlap: a modern polycrisis that runs long enough, and hard enough, to produce an ancient-style regression.
It also implies what to measure. If the outcome depends on the balance between load and repair, then the question for any year is not simply whether things are getting worse, but whether stress is rising faster than the capacity to absorb it — and whether that capacity is being maintained, spent down or built up. Sections 4 and 5 attempt that accounting with public data; the model in section 3 shows why the balance, rather than either quantity on its own, is what matters.
Collapse is common in history, and recovery is slow. What disappears first is complexity: writing, coinage, cities and long-distance trade.
The historical record offers no example of an industrial civilisation collapsing, but it offers many examples of complex societies losing their complexity, and it is unusually consistent about how long the loss lasted. When the Mycenaean palaces fell around 1200 BCE, the Linear B script fell with them; Greece was without writing until the alphabet was adopted around 750 BCE. When the Antonine Plague struck the Roman Empire in 165 CE, the lead pollution that Roman mining and smelting had deposited in Greenland ice fell sharply, and it did not regain its earlier level for more than 500 years. Egypt's Old Kingdom gave way to the First Intermediate Period, about 125 years of fragmented rule, before the Middle Kingdom reunited the country.
These are not exact equivalents of each other, and none is an exact precedent for a global industrial collapse. The durations below are approximate, and some measure different things: the end of writing in one case, a pollution proxy for economic activity in another, the return of cities to a different region in a third. What they share is a pattern. Complexity was lost quickly, within a generation or two, and regained slowly, over centuries — a pattern the model in section 3 reproduces from its own logic.
How long the regression lasted
The Indus figure measures the return of large cities to South Asia, in the Ganges plain rather than the Indus valley. The Classic Maya southern lowlands, which declined c. 760–910 CE, are omitted because only limited parts of them ever fully recovered.
Episodes and markers
| Episode | Marker | Span |
|---|---|---|
| Egypt | Unified rule: Old Kingdom to Middle Kingdom | c. 2181–2055 BCE |
| Mycenaean Greece | Writing: Linear B to the Greek alphabet | c. 1200–750 BCE |
| Roman world | Lead pollution in Greenland ice regains its Pax Romana level | 165 CE – after 665 CE |
| Harappan cities | Large cities return to South Asia (Ganges plain) | c. 1900 – c. 600 BCE |
| Classic Maya | Royal cities of the southern lowlands | c. 760–910 CE, partial recovery |
~336 yrs
Average lifespan of the past civilisations in Luke Kemp's survey for BBC Future (2019)
200 yrs
Period over which premodern states' risk of termination rises before levelling off (Scheffer et al., PNAS 2023)
0
Industrial, globally integrated civilisations that have collapsed — there is no direct precedent
Three lessons follow. The first is that collapse is ordinary rather than exceptional. Survey after survey finds that complex polities tend to last centuries rather than millennia; a survival analysis of premodern states by Marten Scheffer and colleagues found that their risk of ending rises over their first two centuries before levelling off, a pattern the authors interpret as a gradual loss of resilience. Longevity is the exception that requires explanation, not collapse.
The second lesson is that collapse is mostly a loss of complexity, not of people. Populations fell in many of these episodes, sometimes steeply, but humans persisted; what disappeared were the institutions that coordinated large numbers of people — scribal bureaucracies, monetary systems, long-distance trade, urban provisioning. That is why the Polycrisis Apocalypse is defined by regression rather than extinction, and why the speed of recovery depends on how much knowledge and how many institutions survive the fall.
The third lesson is the asymmetry of time. In every case above, the descent took decades and the recovery took centuries. Modern societies are far richer, more knowledgeable and more connected than any of these, which should make them harder to topple and faster to rebuild. But global integration also means there is no outside world from which a collapsed region can re-import complexity, as Western Europe eventually did from Byzantium and the Islamic world. A global collapse would have to be rebuilt from within.
A deliberately simple model of the race between stress and repair. Move the sliders to see when it holds, when it breaks and how long the dark lasts.
The model has two quantities. Capacity, from 0 to 1, stands for a civilisation's complexity and its ability to manage problems: institutions, infrastructure, finance and cooperation. Stress stands for the unresolved load of crises. Each year a crisis may arrive at random and add stress. Capacity repairs stress, but repair saturates when stress is high — the "beyond humanity's capacity to manage" of the definition. Crises reinforce one another in proportion to how interconnected the system is. And unrepaired stress erodes capacity, while capacity regrows slowly, especially from a low base, because rebuilding complexity needs complexity.
Those few assumptions are enough to produce the behaviour the definition describes: long stretches of stability, a threshold beyond which stress runs away and capacity collapses within a few decades, and a dark age of centuries before capacity recovers. The chart shows a single run of 1,000 years; the tiles summarise 300 runs at the same settings; and the heatmap shows how the share of runs collapsing within 300 years depends on repair and coupling together. Everything is computed in your browser, with fixed random seeds so the results can be reproduced.
One simulated millennium
—
of 300 runs collapse within 100 years
—
collapse within 1,000 years
—
median dark age, runs that collapse and recover
—
of all simulated years spent collapsed
The tipping ridge: share of runs collapsing within 300 years
At its default settings — two major crises a decade, moderate coupling and a repair capacity of 0.6 — the model sits close to its threshold. About one run in eight collapses within a century, 45 per cent within 300 years and 91 per cent within 1,000. Among runs that collapse and recover within the millennium, the median dark age lasts 471 years. These settings are not calibrated to the real world; they were chosen to place the model near its edge, where its behaviour is most informative.
The heatmap shows why the balance matters more than either term alone. The boundary between safety and collapse runs diagonally: stronger coupling can be offset by more repair, and weaker repair by looser coupling. Near the ridge, small changes have large effects. With repair at 0.6, moving coupling from 0.22 to 0.28 raises the share of runs collapsing within 300 years from 9 to 74 per cent. This is the signature of a threshold system, and it is why a world that has coped with every shock so far can still be closer to the edge than its record suggests.
The model also separates two kinds of resilience. The levers that change the probability of collapse are on the stress side and the repair side: fewer or smaller crises, looser coupling, more repair. The rebuild setting barely changes the odds of falling, but it changes how long the fall lasts: at the slowest setting runs spend about half of all simulated years collapsed, at the fastest about a fifth. Preventing collapse and shortening it are different jobs, and a serious strategy needs both.
One indicator of stress for each of the ten crisis domains in the definition, with its direction of travel.
The model's lesson is that stress must be read against repair, and both against their trends. The ledger below takes the first step for each of the ten domains in the definition: a current, published indicator of stress, its latest value and whether it is moving the wrong way. It is not a composite index; the domains are measured in incompatible units and adding them would hide more than it showed. Its purpose is to show breadth — how many fronts are under pressure at once — which is the condition for the reinforcement the definition describes.
The charts beneath the table take four of these indicators back in time, because a level means little without a trajectory. Planetary boundaries have been crossed one after another since the first assessment in 2009; the number of state-based armed conflicts has reached its highest level since 1946; global temperature has set successive records; and the material footprint of the average person has risen by almost half since 1990. Section 5 turns to the other side of the ledger: the capacity to repair.
| Domain | Indicator | Latest | Direction | Source |
|---|---|---|---|---|
| Climate & ecology | Planetary boundaries breached | 7 of 9 (2025) | ▲ from 3 in 2009 | Planetary Health Check |
| Resource depletion | Material footprint per person | 12.7 t (2023) | ▲ from 8.7 t in 1990 | UNDP HDR data |
| Water scarcity | People without safely managed drinking water | 2.1 bn (2024) | ▼ coverage 50% → 60% since 2015 | WHO/UNICEF JMP 2025 |
| Food insecurity | People facing hunger | 673 m (2024) | ▼ slightly, above pre-pandemic | FAO SOFI 2025 |
| Economic instability | Low-income countries in or at high risk of debt distress | 32 of 68 (Mar 2026) | ▲ high and persistent | IMF LIC DSA list |
| Geopolitical conflict | State-based armed conflicts | 65 (2025) | ▲ most since 1946 | UCDP/PRIO v26.1 |
| Demographic shifts | Projected peak of world population | ~10.3 bn (mid-2080s) | ◆ ageing, peak this century | UN WPP 2024 |
| Infrastructure decay | Average annual loss to infrastructure and buildings from hazards | $732–845 bn | ▲ rising with climate risk | CDRI 2023 |
| Technological disruption | Doomsday Clock (nuclear, climate, biology, AI) | 85 sec (2026) | ▲ from 100 sec in 2020–22 | Bulletin of the Atomic Scientists |
| Pandemics | Excess deaths associated with COVID-19, 2020–21 | 14.9 m | ◆ largest in a century | WHO |
Planetary boundaries breached, by assessment
State-based armed conflicts, 1946–2025
Global temperature, 1880–2025
Material footprint per person, 1990–2023
Six of the ten domains are moving the wrong way, three are mixed and one is improving. The improving one is instructive: safely managed drinking water reached 60 per cent of the world's population in 2024, up from 50 per cent in 2015, a reminder that sustained investment does move indicators. Hunger fell slightly in 2024 to 673 million people but remains above its pre-pandemic level, and the 2.3 billion people facing moderate or severe food insecurity are a measure of how little slack the food system has.
The worsening domains are also the ones most tightly coupled to others. Breaching planetary boundaries raises the frequency and size of climate and ecological shocks; conflict destroys infrastructure, displaces people and diverts budgets; debt distress removes the fiscal room to respond to anything else. In the language of the model, these are increases in crisis frequency, crisis size and coupling at the same time — the combination that the heatmap shows moving a system toward its ridge.
None of this establishes that the world is on the path the definition describes. The ledger measures stress, and the outcome depends on stress relative to the capacity to repair it, which grew enormously over the past century. The question that matters is whether that capacity is still growing as fast as the load — and the next section suggests that, on several measures, it is not.
Systems approaching a threshold recover more slowly from each shock. Some measures of the world's repair capacity are now slowing or shrinking.
Complex systems often give warning before they tip. As a system loses resilience, it takes longer to return to its normal state after each disturbance, a phenomenon known as critical slowing down. The Pueblo societies of the American Southwest showed exactly this before their transformations, according to a 2021 study by Scheffer and colleagues: their construction activity recovered more and more slowly from climatic shocks in the decades before each collapse. The equivalent question for today is whether the world's capacity to recover from shocks is holding up as the shocks multiply.
Two kinds of evidence bear on it. The first is the pace of human development, which summarises health, education and income, and which slowed sharply after 2019. The second is the money that funds repair across borders — development aid and humanitarian relief — which fell steeply in 2025. Against both should be set the strongest evidence of modern resilience: global life expectancy, which fell during the pandemic and was back above its pre-pandemic level by 2022.
The pace of human development is slowing
But recovery from a shock is still fast
−23.1%
Official development assistance in 2025, real terms: $174.3 bn against $215.1 bn in 2024
50.4%
Share of 2024 humanitarian requirements funded ($24.9 bn of $49.4 bn)
$23 bn
The 2026 "hyper-prioritised" humanitarian appeal, aimed at 87 million people
2022
Year global life expectancy regained its pre-pandemic level (73.0 against 72.9 in 2019)
The global Human Development Index rose by 5.6 thousandths a year in the 2000s and 4.2 in the 2010s. Between 2019 and 2023 it rose by 2.75 a year, after the first recorded fall in 2020; and UNDP projects that the gain for 2024, leaving aside the crisis years, will be the smallest since its series began in 1990. The gap between the lowest and highest development groups, which had been narrowing, has widened again since 2020. This is not collapse, but it is the shape of a system that is recovering less and less fully between shocks.
The finance of repair is shrinking faster. Official development assistance fell by 23.1 per cent in real terms in 2025, the largest annual drop on record, with humanitarian aid falling by more than a third. Humanitarian appeals were half funded in 2024, and the 2026 appeal was cut back to a "hyper-prioritised" $23 billion aimed at 87 million people. In the model's terms, these are reductions in repair capacity at the moment that crisis frequency and coupling are rising — the combination that moves a system toward its ridge from both sides at once.
The counter-evidence is real and should not be understated. Global life expectancy fell from 72.9 years in 2019 to 71.2 in 2021 and was back to 73.0 by 2022, a recovery that would have taken a generation in most of history. Modern capacity remains vast. The early warnings are not that it has failed, but that it is being drawn down and under-replenished — which is how resilience is lost, gradually and then suddenly.
Four levers keep repair ahead of stress. The model shows that each helps on its own and that together they compound.
Because the Polycrisis Apocalypse is defined by a race, every off-ramp is a way of changing one of its terms. Fewer and smaller crises reduce the load. Looser coupling stops one crisis from feeding the next. Greater repair capacity resolves stress before it accumulates. And faster rebuilding shortens any regression that does occur. The four cards below translate each lever into the model's terms and into the kinds of policy that move it in the real world; the figures are the model's, at default settings, and are illustrative.
The most important result is that the levers compound. Each on its own cuts the share of runs collapsing within 1,000 years from 91 per cent to between 10 and 28 per cent. Together — slightly fewer crises, slightly looser coupling and a little more repair — they remove it entirely in the model's 300 runs. Near a threshold, modest improvements on several fronts are worth more than a heroic effort on one.
LEVER 1 · LOAD
Reduce how often crises arrive and how large they are, at their source: mitigation that slows warming, prevention that stops outbreaks spreading, diplomacy that prevents wars.
91% → 26%
Model runs collapsing within 1,000 years when crises arrive half as often
LEVER 2 · COUPLING
Stop one crisis from feeding the next: keep trade open in emergencies, diversify supply, hold buffer stocks and separate critical systems so a failure in one does not cascade.
91% → 10%
Model runs collapsing within 1,000 years when coupling falls from 0.25 to 0.20
LEVER 3 · REPAIR
Sustain and grow the capacity to resolve stress: state capacity, fiscal space, debt relief, adaptation finance and the international institutions that pool risk across borders.
91% → 27%
Model runs collapsing within 1,000 years when repair capacity rises from 0.6 to 0.7
LEVER 4 · REBUILD
Shorten any regression by keeping knowledge, institutions and seed stock intact: archives, distributed expertise, local capacity and the foundations of science and education.
51% → 19%
Share of simulated years spent collapsed, slowest against fastest rebuild setting
The model's levers map onto debates that are already live. Shrinking the shocks is the domain of mitigation and prevention, which pay off slowly and are perennially underfunded relative to response. Loosening the links is the logic of trade rules, strategic reserves and supply-chain diversification, which the food and energy shocks of the 2020s made concrete. Funding the repair is the agenda of debt relief, development finance and adaptation, which is moving backwards in 2025 and 2026.
The fourth lever is the least discussed and perhaps the most distinctive to this concept. Because a Polycrisis Apocalypse is defined by the length of its regression, anything that preserves the means to rebuild — knowledge, institutions, genetic resources, skills held locally as well as globally — directly reduces its harm, even if it does nothing to prevent the fall. Past recoveries were faster where literacy, law or trade survived somewhere; the libraries and monasteries that carried texts through the European early Middle Ages are the historical form of this lever.
The conclusion is not that collapse is imminent, and nothing in this dashboard supports a date or a probability. It is that the outcome the definition describes is the product of a race whose terms can be measured and changed, that several of those terms are moving the wrong way at once, and that the cheapest moment to change them is while the system is still on the safe side of its ridge. The concept is useful precisely because it names what is at stake if that moment is missed.
How the model works, what it assumes and what it cannot tell you.
The spiral model runs in annual steps. Each year a crisis arrives with probability equal to the chosen frequency per decade divided by ten; its size is drawn from an exponential distribution with the chosen mean and scaled by 0.3 + 0.7 × capacity, so that simpler societies face smaller systemic shocks. Stress then changes by the crisis, plus coupling × capacity × stress, minus repair × capacity × stress / (1 + stress / 0.5), minus 0.03 × stress. Capacity changes by 0.2 × capacity × (1 − capacity) × capacity / (capacity + a), minus 0.06 × stress × capacity, with a floor of 0.02. The rebuild setting sets a to 1, 0.5, 0.3, 0.15 or 0.05.
A run counts as collapsed when capacity falls below 0.25 and as recovered when it next rises above 0.75. The tiles use 300 runs with seeds 1000 to 1299 over 1,000 years; the heatmap uses 100 runs per cell, seeds 1000 to 1099, over 300 years. The model was written and calibrated in Node.js and the browser runs the identical code, so the figures quoted in the text can be reproduced on the page. Parameters were chosen to place the default near the model's threshold, not estimated from data; the model's outputs are therefore illustrative of mechanism and say nothing about the real probability or timing of collapse.
Historical durations are approximate, drawn from the sources below, and measure different markers. Present-day indicators are the latest published values at the time of writing, from the sources listed; the HDI growth rates and the material-footprint and life-expectancy series are the author's calculations from the UNDP Human Development Report 2025 data. Related analysis in this series: the polycrisis century measures how often systems are stressed at once, and resilience in an age of polycrisis traces how a single shock cascades and what amplifies it.
Data and literature consulted in October 2026.