A framework hypothesis

Kari McKern ·  · 4 October 2026 · framework hypothesis paper

Abstract

Human organisation is the coordination system that lets primates who will never meet share skills, surpluses and a common picture of the world. The Complex Adaptive Model of Societies (CAMS) treats that system as a group adaptation to a setting, not as a moral agent. The framework reads a society through eight functions, the Paladins — Helm, Shield, Lore, Archive, Craft, Flow, Stewards and Hands — each carrying two coordinates. Shared symbolic organisation is P=CA, Coherence times Abstraction. Functional headroom is H=K-S, Capacity minus Stress. The pattern of coupling between the Paladins is the society's topology. The Zeitgeist is the readable surface of that topology.

The framework itself is a frame, in the sense that natural selection is a frame. It is not directly falsifiable, and it is not offered as one. What follows from it is. This paper states six hypotheses, each with a test and a result that would retire it. A blind deep-time study of 2 October 2026 is reported as a first look, not a confirmation. One pre-registered hypothesis from that study, H4, failed and is reported as a failure.

1. The frame

A band of forty can coordinate by knowing one another. Past a few hundred, trust has to become symbolic. Strangers have to agree on a price, honour a debt, obey a ruling, and believe that grain handed over this season will still be accounted for next season. Every civilisation is a solution to that problem.

Societies are pods: evolving primates in different thermodynamic settings, with different strategies for making a living. In Shark Bay, some bottlenose dolphins carry sponges on their beaks to forage the seabed. Off Patagonia, orcas surge onto beaches in seal-pup season. Humpbacks blow bubble-nets. Nobody calls the sponge-carriers lazy or the orcas evil. Each group has adapted to its setting. Human societies are read the same way. A river valley that must move earth and hold water grows a bureaucracy of account and devolution. A coast grows boats, and wealth that can be carried. An open continent grows buffers. A steppe grows speed. Aboriginal Australia and Māori Aotearoa were close to optimal for their settings and technologies. None of these is a rung on a ladder.

Complexity driven hard produces new layers. Money, citizenship, property, the corporation, the nation, rank, law, debt and borders are abstractions. None can be touched. Once enough people coordinate around them, they constrain the people who reproduce them. The higher-order entity wants nothing. It is real because changing it requires coordinated action at the level below. Being a primate is also about managing aggression. Under surplus, deliberation is affordable. Under strain, patience shortens and force becomes cheap. A theory of human organisation that leaves that out is a theory of some other animal.

2. The instrument

The eight Paladins are functions, not classes and not persons. Helm, at the Throne, decides what happens next for people who will never meet. Shield, on the Battlefield, keeps force and order. Lore, in the Temple, makes the meaning. Archive, in the Library, keeps the record so strangers can trust the same numbers. Craft, in the Workshop, turns abstraction into things. Flow, in the Agora, keeps goods moving. Stewards, in the Manor, hold the land and the stores. Hands, in the Fields, move the earth. Three make headroom: Hands, Craft and Flow. Two keep the shared models: Lore and Archive. Two are switches: Helm and Shield. Stewards stand between the makers and the keepers.

Each Paladin i is scored from 1 to 10 on Coherence Ci, Capacity Ki, Stress Si and Abstraction Ai, under the JUNO v1.2 rubric. This paper reads those four numbers as two coordinates:

[ P_i = C_i A_i, \qquad H_i = K_i - S_i ]

Pi is shared symbolic organisation: how much common model a function carries. It is not intelligence. Hi is functional headroom: the margin between what a function can presently do and what is demanded of it. Summed, the first coordinate is

[ \Psi = \sum_{i=1}^{8} C_i A_i ]

Two societies with the same (\Psi) can arrange their eight points differently. The shape is the object of study. The Paladins are coupled. Let Wij be the coupling recovered from how they move together, and let headroom set the gain:

[ W_{ij}^{*}(t) = W_{ij}, g\bigl(H_i(t), H_j(t)\bigr) ]

The relationship persists. Surplus, threat, exhaustion and disruption decide which connections dominate. A society's topology is that pattern. Its dominant myth is the story that justifies the coupling it is already leaning on: Throne with Temple, Battlefield with Temple, Manor with Library, Agora with Workshop, Fields with Temple. A society does not choose the myth freely. It falls into the one its topology can carry.

Under surplus the Paladins can afford to differ. Under sustained negative headroom, coupling concentrates along Helm, Shield and Lore. That is constriction: organisation reconfigured for threat, not a loss of organisation. Constriction is fast. Re-differentiation is slow. If coupling stays tight after headroom returns, the society remembers the excursion. That is hysteresis, and it is measurable at matched headroom on the way into and out of a crisis.

The frame predicts two kinds of collapse. In a decoherence-first collapse, shared symbolic organisation fails at the top while headroom is still positive. Recovery can be fast, because the base survives. In a headroom-first collapse, a material shock strikes a system whose top still looks coherent, and takes the stores of shared models down with the base. Recovery is slow.

CAMS measures how strongly functions hang together and how much margin they have. It does not measure virtue. Germany's score rises in 1933. Iran's rebounds in 1954. Health, legitimacy and democratic capacity are the wrong names for the field.

3. What the frame is not

The frame is not a ranking, a forecast, or a moral theory. Predictive power on present panels is modest: the honest pooled figure is an AUC of 0.563, where 0.5 is chance. CAMS is a thermometer. Whether it forecasts is what the locked register and the six hypotheses are for.

The names are recent. The functions began as descriptive roles and took their present names under JUNO v1.2, in wording that privileges no one culture. The number and the cuts are a scaffold for scoring, not a discovery about nature.

4. Six hypotheses

Each hypothesis is logged with its test, threshold and failure condition before new scoring. Topology, attractors and Paladin readings are computed only after the scores are sealed. Failures are published as failures.

H1. Topology lasts longer than regimes. A society's coupling pattern survives changes of government, dynasty and ideology, and changes slowly with its setting and livelihood. It fails if the coupling reshuffles with each change of regime as much as it differs between unrelated societies.

H2. Constriction comes before force. Under sustained negative headroom, coupling concentrates along Helm, Shield and Lore, and the range of responses narrows, before the society turns to large-scale coercion or war. It fails if the concentration appears only at or after the turn to force. Constriction would then be a symptom, not a mechanism. The test needs event data from outside CAMS.

H3. Recovery is slower than the fall. At equal headroom, a society recovering from a crisis has a different topology from the same society on its way in. It fails if the way out retraces the way in at matched levels of K-S.

H4. Two kinds of collapse. Decoherence-first collapses recover faster than headroom-first collapses that destroy the stores of shared models. It fails if recovery rates do not differ by collapse type across further series: Ur III, the Shang–Zhou transition, the Late Bronze Age Levant, and the post-Harappan Indus. The Egypt / Aegean contrast below is an illustration, not this test.

H5. Myth follows coupling. The dominant myth of an era binds the same pair of functions that the scores show as most tightly coupled. It fails if myths coded blind from texts — speeches, sermons, editorials, inscriptions — by coders who have not seen the scores move independently of the coupling recovered from those scores.

H6. The setting predicts the topology. River-valley, maritime, continental and steppe societies show their characteristic couplings across unrelated cultures. It fails if the setting predicts coupling no better than chance once the famous cases are set aside.

If one hypothesis fails, the corresponding part of the mechanism is wrong. If most fail, the frame can still describe societies but not explain them.

5. First look, not confirmation

On 2 October 2026 Neural Nations published Reading Societies Without Their Stories, a blind, pre-registered test of CAMS on ancient collapse and recovery. Three model families scored twenty ancient society-windows under the v1.2-OPT rubric with a deep-time adapter, in 22 blind passes per prompt: 15 from Claude, 4 from Grok and 3 from GPT. A node was scored only if the scorer could name evidence of performance, of strain, and of a mechanism linking them. Hypotheses were logged before scoring. Passes that broke blindness were excluded.

The two collapse types were articulated after that study. Its results are a first look.

In Egypt, mean Coherence of Helm and Lore fell across the two windows before the First Intermediate Period while headroom was still positive: from 8.1 to 5.1 under Claude, 8.0 to 5.4 under Grok, and 8.0 to 6.5 under GPT. In the Aegean it stayed flat until the breakdown. Egypt's collapse began as a loss of shared purpose at the top. The Aegean's struck a system that still looked coherent. The Aegean fall was the steepest in the data, 4 to 7 points of headroom per century, with recovery at 1.3 to 2.0. Under Claude and Grok the Aegean was still below its pre-collapse level 500 years later. Egypt rebuilt toward Old Kingdom levels in under three centuries. Linear B vanished with the palaces. Egypt kept its scribes. These series were scored in Node Value and in the underlying C, K, S and A. They have not yet been recomputed as P and H. Until they are, they illustrate the distinction. They do not establish H4.

Lore held best in both collapses. Hands were the lowest or second-lowest node in almost every state window. The non-state societies in the study, Aboriginal Australia and Māori Aotearoa, did not show that gap. Maritime societies had Flow above Helm. A separate pre-registered hypothesis, that Mesopotamian debt-relief edicts kept labour stress lower for longer, failed under every model. Hands Stress was below the comparator in none of the three windows. The design is flawed, because edicts were issued because labour was already under debt strain. It is reported as a failure, because that is what the pre-registered test showed.

The models ranked the ten time windows almost identically. These are among the most written-about collapses in archaeology. Agreement shows consistency, not historical truth. No human specialist has yet scored a case. GPT did not accept oral tradition, and left Lore and Archive unscored for every society without writing. Claude and Grok scored Aboriginal Lore and Archive at 7 to 8. The next deep-time rubric, DT-2, states that oral tradition is evidence for Lore and Archive, separates "function demonstrably absent" from "no evidence", and adds guidance on Abstraction for coerced but skilled labour.

6. The next test

Six deeply separated cases, each a different solution to the coordination problem: a mobile forager society; Liangzhu, lower Yangtze, c. 3300–2300 BCE; Old Kingdom Egypt; Mycenaean Pylos; Qin and Han China; a maritime Polynesian society. Egypt and Pylos were already in the 2 October study. The others are new. Each case is scored as a series of windows where the record allows.

The order is the protection. First, score all eight Paladins under the locked DT-2 rubric, without the cognitive interpretation and without the hypotheses in view. Run several model families blind, add a fourth model, and seal the scores. Before unsealing, log the expected tight pair for each society under H6. Second, calculate Pi, Hi, (\Psi), coupling and attractor, and where a series crosses a crisis compare topology at matched headroom. Third, bring in archaeologists and historians in two or three regions, scoring their own cases on the same rubric. Fourth, and last, reconstruct what it was like to stand in each place. Those voices are inference from the locked scores, labelled as such, and they stay outside the evidence.

If the resulting landscapes recover nothing beyond what entered the scores, the hypothesis dies.

7. Limits

Every CAMS scorer so far is a language model, and the models share a broad family of training priors. Hiding the names of societies does not hide history from scorers trained on that history. The panels are dominated by one shared field, and structure beyond the first two factors lies within about twice the disagreement between scorers. Whether P and H vary independently enough to separate the four regimes is itself a question for the data. Claims H2 and H5 need evidence from outside CAMS: event datasets for the timing of coercion and war, and myth coded from texts by people who have not seen the scores.

8. Standing

Darwin's principle, that what fits its setting persists, is not directly falsifiable. It earned its standing because it generated claims that could have failed. CAMS has the same structure. The frame does not have to be falsifiable. The six hypotheses do. Specialists are invited to rescore or contest any case. Data, prompts and code for the deep-time study are in the backstory folder of the KaliBond/wintermute repository.

Kari McKern, Neural Nations /Framework hypothesis paper, 4 October 2026.