Can We Read the Warning Signs of Societal Collapse?

A new study tracks eight markers of institutional health across 44 countries and 125 years — and correctly identifies which political crises were genuine structural collapses, and which were simply change.

In 1997, Hong Kong passed from British to Chinese control in one of the most watched political transitions of the late twentieth century. Protests were small. Markets held. The civil service stayed in place. By any reasonable measure, it was a high-tension moment in a city's history — but not a collapse.

Twenty-two years later, in 2019, something different happened. Mass protests, tear gas, arrests, a security law that dismantled civil liberties. The institutions that had held the 1997 transition together began to fracture.

What if a model could tell the difference — in advance?

A new cross-national study suggests that it can. Using a framework that scores the health of eight core institutional functions — from governance and legal memory to economic capacity and information flow — researchers found that the structural signature of genuine societal collapse is measurably different from the signature of political transition. And across 44 countries and 125 years of history, that signature predicted collapse with an accuracy of 74 percent.

What the framework measures

The study uses a system called CAMS — Complex Adaptive Meta-System — which treats any functioning society as a network of eight interdependent institutions: executive governance, security and defence, normative culture, resource stewardship, institutional memory and law, technical capacity, labour and production, and information exchange. Each institution is scored annually on four dimensions: how coherent it is, how capable, how stressed, and how capable of generating new ideas.

From those scores, the framework derives a key quantity called expressed cognition — essentially, whether institutions are able to translate their knowledge and organisation into effective action. When stress outstrips capacity, expressed cognition inverts: the system's knowledge turns inward, becoming reactive and self-protective rather than generative. This is the signature the study was looking for.

"What precedes collapse is not dramatic chaos in the visible functions of society. It is the quiet erosion of the slow nodes — the institutions that hold memory, legal authority, and governance capacity — while the faster, operational systems continue to run."

Telling collapse from change

The study's most important finding came from a distinction that sounds simple but turned out to be analytically decisive: not all political ruptures are the same.

The researchers coded 126 major political events across the study panel — coups, civil wars, revolutions, economic crises, independence movements, regime changes. When they fed all of these into the model, the accuracy was moderate: about 65 percent. Reasonable, but not impressive. Then they looked more carefully at what was being miscategorised.

What they found was that the model was consistently correct about one type of event that it was being penalised for getting right. Events like Hong Kong's 1997 handover, Estonia's 1991 independence, Finland's 1918 civil war (which ended in rapid reconstruction), and several coups that transferred power without systemic collapse — these all registered in the framework as coordination transitions, not collapses. High organisational energy, strong institutional coupling, positive expressed cognition. The model assigned them low collapse probability. The historical record agreed. But they were coded as "ruptures" and counted as prediction failures.

When the researchers separated genuine structural collapses — events where expressed cognition went deeply negative, institutional bonds weakened, and the capacity to coordinate under stress collapsed — from these coordination transitions, the picture changed entirely. On structural collapses specifically, the model's accuracy rose to 74 percent across 29 countries with evaluable histories. In 27 of those 29 countries, the model outperformed chance.

The mechanism: when symbols stop working

The study draws on a parallel body of theoretical work about how human societies function as extended primate herds — systems where trust is maintained not through direct social contact, but through symbols, institutions, and shared narrative. Laws, ceremonies, official communications, and institutional guarantees all function as what the researchers call "symbolic grooming" — they regulate collective stress by conveying that the group will absorb some portion of each member's risk.

This works until it doesn't. When institutions produce the language of care without the substance — when governments announce consultations that are already decided, when employers celebrate wellbeing while cutting capacity, when legal frameworks multiply while enforcement weakens — the signal of trust is emitted without the cost being paid. Stress remains unresolved. People learn to distinguish the performance from the reality.

The structural signature of this failure shows up in the CAMS data as a widening gap between institutional nodes that produce legitimacy narratives and those that deliver material outcomes. Legal memory erodes. Governance capacity declines. The coupling between what institutions say and what they do weakens. Institutional bonds — the measure of how tightly the eight functions reinforce each other — fall. Stress rises.

Eventually, the system reaches a threshold where small shocks produce disproportionately large responses. The herd, which has been watching for signs of flight for years, begins to move.

Where things stand now

Three countries in the study panel are currently at or above their own all-time stress readings within the 125-year dataset:

South Africa's institutional memory node — the part of the framework that encodes legal precedent, administrative continuity, and the system's capacity to learn across governments — collapsed ten points in five years (from a healthy 7.5 in 2020 to −2.5 in 2025). The only parallel in the 125-year dataset is Afghanistan after 2001. The United States reading is the more counterintuitive finding: CAMS stress in 2025 is higher than in any year of the Great Depression, measuring the structural coordination capacity of institutions rather than economic output alone.

What an accuracy of 74 percent means

A 74 percent accuracy rate sounds modest until you consider what the model is doing: it is looking at eight institutional scores for a country in a given year, with no access to news headlines, economic indicators, political polling, or any other conventional input, and predicting whether that country will experience a major structural collapse within the next three years. It is doing this across every continent, every political system, and every decade from 1900 to the present.

Political scientists who build early-warning models with far richer inputs — income data, democracy indices, ethnic composition, prior conflict history, election results — typically achieve accuracies in the 68–75 percent range, and those models are tested on more narrowly defined outcomes over shorter time horizons. The CAMS framework is in the same range using only internal structural measures.

More importantly, the 74 percent figure applies specifically to the events the framework is theoretically designed to predict: cases where the system's capacity to coordinate under stress has genuinely collapsed, not cases where power changed hands smoothly. When the model says a society is at high structural risk, it has historically been right in three out of four cases.

The India exception

In 42 of 44 countries, stress suppresses expressed cognition — when institutions are under pressure, their capacity to translate knowledge into action declines. The one clear exception is India, where the relationship runs in the opposite direction: higher stress is associated with higher expressed cognition. This may reflect India's particular institutional history, in which periods of external pressure or internal tension have sometimes mobilised rather than suppressed institutional energy. It is a genuine anomaly worth investigating rather than explaining away.

What comes next

The study establishes that the structural approach works when applied to the right question. What it has not yet tested is whether adding measures of narrative coherence — the degree to which a society's occupational and institutional stories remain connected to its material conditions — improves prediction beyond the structural baseline. That test is the next step.

The theoretical proposition is clear: civilisations depend on their capacity to extend primate trust through symbols. When those symbols stop connecting to material reality — when the language of care is emitted without the cost being paid — the herd notices. What the data now suggests is that the moment of noticing is detectable, measurable, and precedes the visible break by years.

The monkeys can tolerate a great deal. What they cannot tolerate indefinitely is being groomed by a hand that never touches them.