Kari Freyr McKern Complex Adaptive Humans
Draft — March 2026
This paper presents the Complex Adaptive Model of Societies (CAMS), a framework derived from first principles of multi-timescale coordination in complex adaptive systems. CAMS models civilisations as far-from-equilibrium metabolic networks composed of eight invariant functional nodes, each characterised by four state variables. The eight-node architecture is not an empirical taxonomy but a deductive consequence of stability requirements: models with fewer nodes force incompatible timescales into single variables (compression cost), while models with more nodes face quadratic coupling explosion (fragmentation cost). Eight represents the minimal partition preserving both functional distinctness and tractable inter-nodal synchronisation — four fast-loop nodes for entropy export and four slow-loop nodes for negentropy maintenance.
The framework was derived theoretically, then tested empirically. Cross-societal validation across 18 societies spanning approximately 2,000 years of institutional history (from Imperial Rome to contemporary nation-states) reveals strikingly invariant coordination dynamics: universal stress-capacity anti-correlation (ρ = −0.66, N = 25 dataset-assessor combinations, 25/25 negative), near-perfect bond-strength-to-system-health coupling (mean ρ = +0.93, 24/24 positive), and convergent structural signatures across independent AI assessors (mean cross-LLM concordance ρ = 0.66–0.73 on key metrics).
The central theoretical result — the Coordination Phase Transition Theorem — establishes that civilisational crisis is a coupling-mediated desynchronisation event, not a product of individual node failure or agentic incompetence. This result is grounded in the Kuramoto synchronisation literature and the Master Stability Function formalism: the criticality index 𝒳(t) = D_Ψ(t)/Λ(t) is structurally identical to the disorder-to-coupling ratio governing phase transitions in coupled oscillator networks. Algebraic connectivity of the coordination Laplacian provides a computable, falsifiable diagnostic for synchronisation capacity.
The clarity and universality of the dynamics observed in the data have driven successive refinements of the theoretical apparatus. CAMS represents the current stage of that iterative process: a formal model adequate to the empirical signal.
Every complex society faces the same fundamental engineering problem: how to synchronise institutions operating on radically different timescales. A logistics network responds to disruption within weeks. A legal tradition evolves over decades. A labour market adjusts seasonally. A cultural mythos shifts across generations. When these timescales desynchronise — when fast-loop institutions outrun the slow-loop structures that give them coherence — the result is not merely political difficulty but a structural phase transition in the coordination state of the system.
This paper argues that the architecture required to manage this multi-timescale coordination problem is neither arbitrary nor culturally contingent. It is constrained by the mathematics of coupled dynamical systems operating across incommensurate rates.
CAMS was not constructed by observing societies and inductively categorising their institutions. The derivation followed a different sequence:
The striking feature of this process was not that the theory was confirmed — any sufficiently flexible model can be fitted to data. The striking feature was the clarity and universality of the dynamics the data revealed. Across liberal democracies, social democracies, theocratic republics, authoritarian states, and pre-modern empires, the same coordination physics emerged with remarkable consistency. The theoretical development since the initial derivation has been a process of building mathematical apparatus adequate to describe dynamics that were, empirically, far more regular than anticipated.
This paper makes three claims, in order of strength:
Claim 1 (Architectural). The eight-node partition is the metastable optimum for multi-timescale societal coordination, derivable from rate-separation and coupling-cost arguments without reference to any specific society.
Claim 2 (Dynamical). Civilisational crisis is a coordination phase transition — a structural desynchronisation event characterised by the ratio of inter-nodal rate dispersion to coupling capacity exceeding a critical threshold. This is a property of the system's phase space, not of any individual node's failure.
Claim 3 (Universal). The coordination physics described by CAMS operates identically across governance models, ideological systems, and cultural traditions. Societies classified as geopolitical rivals share more coordination structure than competitive framings suggest.
Large societies exhibit the defining properties of complex adaptive systems: distributed decision-making, feedback loops, adaptive learning, and continuous throughput of energy, materials, and information. However, unlike many engineered networks, societies must coordinate processes operating at very different characteristic speeds.
Two broad dynamical domains are present:
Fast operational loops respond to immediate shocks and material demands: labour execution, logistics and circulation, specialised production, coercive enforcement.
Slow integrative loops stabilise the system across time: memory and record, meaning systems, resource allocation, executive coordination.
Let each institutional process be represented by a state variable x_i with characteristic relaxation time τ_i. Stability of the overall system depends on whether these heterogeneous rates remain synchronised through coupling.
The modelling problem therefore becomes: What is the smallest partition of institutional processes that preserves rate separation while maintaining stable coupling?
If too few nodes are used, distinct functions must be compressed into single state variables. For example: merging coordination and coercion (Helm–Shield), merging ideology and institutional memory (Lore–Archive), or merging mass labour and specialised production (Hands–Craft).
This compression forces processes with different intrinsic response times into the same dynamical compartment. In oscillator terms, the node must simultaneously accommodate multiple eigenmodes with incompatible frequencies. Internal coherence degrades because the node must respond both quickly and slowly to the same stimuli.
Empirically this manifests as unstable state estimates, rapid variance inflation, and loss of predictive discrimination. Architectures with fewer than roughly eight functional partitions exhibit compression instability.
At the opposite extreme, increasing the number of nodes raises coupling complexity. For N nodes, potential pairwise couplings scale as N(N−1)/2. Each coupling channel carries coordination cost: information transfer, conflict resolution, and synchronisation overhead. As node count increases, average coupling strength declines, noise propagates through the graph, and system coherence becomes fragile.
In statistical terms, parameter freedom increases faster than explanatory power. The model begins fitting noise rather than structural dynamics. Architectures with large N suffer from coupling fragmentation.
The competing pressures of compression and fragmentation can be formalised as a standard model-order selection problem. Define:
$$N^* = \arg\min_N \left[ \mathcal{E}(N) + \lambda,\mathcal{C}(N) \right]$$
where ℰ(N) is compression error (decreasing in N) and 𝒞(N) ∝ N(N−1)/2 is coupling/complexity cost (increasing superlinearly in N). This framing connects the eight-node claim to established information-theoretic criteria (AIC, BIC, MDL): the optimal number of nodes minimises the sum of fit loss and model complexity under explicit penalty.
The prediction is clear: sub-eight variants should show aggregation bias (compression artefacts); super-eight variants should show coupling noise and degraded generalisation. This is an adjudicable empirical claim under standard model-selection frameworks.
The viable architecture lies between these two extremes. A minimal stable partition must: (1) preserve separation between fast and slow institutional processes; (2) maintain functional orthogonality between roles; and (3) keep coupling complexity manageable.
When these constraints are applied, a natural partition emerges consisting of two functional quartets.
Fast-loop quartet (entropy export):
Slow-loop quartet (negentropy maintenance):
This fast/slow partition is not a descriptive convenience. It is the structural feature that makes the eight-node model dynamically meaningful: crises arise precisely when the fast quartet outpaces the slow quartet's capacity to maintain coherent coordination.
The architecture reflects a deeper symmetry. A civilisation-scale system must maintain two interacting metabolisms: a material metabolism (act, move, protect, produce) and a symbolic metabolism (remember, interpret, allocate, steer). Each metabolism requires roughly four functional roles, producing the observed 4 + 4 architecture.
The rate-separation argument provides a theoretical explanation for why the eight-node architecture is plausible and positions it as an empirically adjudicable claim under information-theoretic model selection. It does not yet constitute a formal proof. Further work should test whether: (i) models with 5–7 nodes show systematic compression artefacts; (ii) models with >8 nodes increase noise without improving prediction; and (iii) eigenmode analysis confirms eight dominant coordination modes in the 32-dimensional state space.
Let N = {Archive, Craft, Flow, Hands, Helm, Lore, Shield, Stewards} with |N| = 8. For node i ∈ N at time t ∈ ℝ⁺, define the nodal state vector:
$$\mathbf{m}_i(t) = \begin{bmatrix} C_i(t) \ K_i(t) \ S_i(t) \ A_i(t) \end{bmatrix} \in \mathbb{R}^4$$
where C_i(t) is Coherence (cognitive integration), K_i(t) is Capacity (collective efficacy), S_i(t) is Stress (affective load; higher = higher load after polarity normalisation), and A_i(t) is Abstraction (representational complexity — the degree to which the node operates through meaning, narrative, and codified belief, not a gap between self-narrative and material reality).
Define b_i(t) ∈ ℝ⁺ as nodal bond strength (coupling potential). Operationally, b_i(t) is a node-level proxy for how strongly node i participates in system-wide coordination; it serves as both an aggregate weight and an edge-strength modulator in the coupling network W(t).
System state:
$$X(t) = {m_i(t), b_i(t) \mid i \in N} \in \mathbb{R}^{8 \times 5}$$
Node Value (composite health measure):
$$V_i = C_i + K_i - S_i + 0.5 \cdot A_i$$
Social Cognition (per node):
$$SC_i(t) = \frac{A_i(t)}{C_i(t) + \varepsilon_c}$$
Values above 1 indicate abstraction-dominant cognition; below 1 indicate concrete/pragmatic cognition.
National Affect (per node):
$$NA_i(t) = K_i(t) - S_i(t)$$
Positive values indicate buffered capacity; negative values indicate strain.
System aggregates are reported in both intrinsic (unweighted mean) and coupling-effective (bond-weighted) forms:
$$CC_{\text{eff}}(t) = \frac{\sum_i b_i(t) \cdot SC_i(t)}{\sum_i b_i(t) + \varepsilon_b}, \qquad CA_{\text{eff}}(t) = \frac{\sum_i b_i(t) \cdot NA_i(t)}{\sum_i b_i(t) + \varepsilon_b}$$
Cognitive Preservation Under Stress (Σ / "Sisu" operator). A continuous dissociation index:
$$\Sigma_i(t) = \left(A_i(t) - A_0\right)+ \cdot \left(-NA_i(t) - NA_0\right)+$$
where (x)₊ = max(x, 0), with default thresholds A₀ = 7 and NA₀ = 0.5. High Σ indicates high abstraction concurrent with strongly negative affect — the society maintains symbolic complexity while under severe material strain.
Rate Dispersion (the institutional "shear"):
$$\Omega(t) = \text{std}_i(\Delta S_i(t))$$
where ΔS_i(t) = S_i(t) − S_i(t−1). This measures the degree to which nodes are accelerating at different speeds.
Define a time-varying symmetric coupling matrix W(t) ∈ ℝ^{8×8} with W_ii(t) = 0. W(t) is interpreted as undirected coordination conductance (bond channel strength), distinct from directed causal influence.
Construction. Let m(t) denote the collection of nodal states {m_i(t)}. Define:
$$W_{ij}(t) = b_i(t) \cdot b_j(t) \cdot \exp\left(-\frac{|\mathbf{m}_i(t) - \mathbf{m}_j(t)|^2}{2\sigma(t)^2}\right)$$
where σ(t) is set by a median pairwise distance heuristic unless specified. Distances are computed on z-scored components of m within each time slice to prevent any component from dominating the kernel geometry.
Reconstruction rule. W(t) is reconstructed as a deterministic function of current nodal states and coupling potentials. In empirical analyses, W(t) is evaluated on observed states; in forward simulation, on simulated states. We do not model dW/dt in the core formulation.
The coupling architecture described in Section 3.3 connects directly to the theory of synchronisation in coupled oscillator networks. This connection is not analogical — it is mathematical.
Kuramoto-type threshold. In the classical Kuramoto model, a population of oscillators with heterogeneous natural frequencies ω_i achieves coherent collective behaviour when coupling strength K exceeds a threshold K_c proportional to the spread of intrinsic frequencies. Below this threshold, oscillators rotate independently; above it, a macroscopic fraction locks into synchrony.
The CAMS criticality index (Section 4.1) has precisely this structure. Rate dispersion D_Ψ(t) plays the role of frequency spread; aggregate bond strength Λ(t) plays the role of coupling strength K. The ratio 𝒳(t) = D_Ψ/Λ is the analogue of the disorder-to-coupling parameter that governs the synchronisation phase transition. Crisis occurs when dispersion overwhelms coupling — exactly the Kuramoto desynchronisation mechanism operating on institutional timescales rather than oscillator phases.
Master Stability Function formalism. For a network of coupled dynamical systems, the Master Stability Function (MSF) determines synchronisability from the spectral properties of the coupling matrix. Specifically, the Laplacian L_W = D − W (where D_ii = Σ_j W_ij) encodes the network's capacity to maintain synchronous evolution. The condition for stable synchronisation depends on the eigenvalue ratio λ_N/λ_2 of the Laplacian, where λ_2 (algebraic connectivity) is the critical quantity.
This provides a principled, computable diagnostic for CAMS: algebraic connectivity a₂(t) = λ₂(L_W(t)) measures the coordination network's synchronisability at each time step. The prediction is that a₂ degrades prior to crisis events — the coordination graph becomes structurally less synchronisable before the rupture manifests in node-level state variables.
Node evolution via graph diffusion. The coupling influence on node dynamics takes the standard Laplacian diffusion form:
$$\Psi_{\text{coup},k}(i) = -\delta_k (L_W m_k)_i$$
where m_k denotes the k-th component field over nodes (for k ∈ {C, K, S, A}). This is diffusive coupling: nodes are pulled toward their neighbours proportional to edge conductances. Combined with OU-like internal drift and optional external forcing, the full discrete-time update is:
$$m(t+1) = m(t) + \Psi_{\text{int}}(m(t)) + \Psi_{\text{coup}}(W(t), m(t)) + \Psi_{\text{ext}}(t) + \varepsilon(t)$$
Bond dispersion (heterogeneity): σ²(t) = Var_i(b_i(t))
Bond profile persistence (rank-order inertia): ρ_profile(t) = Corr(b(t), b(t−1))
Crisis risk score (state-dependent, not a calibrated probability). Using normalised bonds b^norm_i(t) ∈ [0,1]:
$$S_{\text{crisis}}(t+\Delta) = \sigma\left(\theta_0 + \theta_1 \sigma^2(t) + \theta_2 \rho_{\text{profile}}(t) - \theta_3 \bar{b}(t)\right)$$
where σ(·) is logistic sigmoid and Δ = 2 is the default lead time.
For a node-level metric series x_i(t), define the autocorrelation function R_i(τ) = Corr(x_i(t), x_i(t+τ)). Stationarity is assessed via ADF test with constant only; if a unit root is not rejected at 5%, the series is first-differenced.
$$\tau_{1/2,i} = \min{\tau : |R_i(\tau)| \leq 0.5}$$
Nodes are ranked by τ_{1/2} to characterise institutional memory structure.
Directed influence is defined separately from undirected coupling. The causal strength tensor C ∈ ℝ^{8×8×L} uses block Granger causality: a VAR on the concatenated vector [m_j(t), m_i(t)] ∈ ℝ⁸ tests whether the lagged block of m_i jointly improves prediction of the target block m_j (block F / Wald test).
Influence centrality (driver strength): IC_i = Σ_j Σ_ℓ C_{ij}(ℓ)
Receptivity (responder strength): IR_i = Σ_j Σ_ℓ C_{ji}(ℓ)
Let f denote the one-year map m(t+1) = f(m(t)). Linearised stability at state m* is determined by the spectral radius of the Jacobian:
$$\rho(J) = \max_i |\lambda_i(J)|$$
where ρ(J) < 1 indicates local asymptotic stability, ρ(J) ≈ 1 indicates criticality, and ρ(J) > 1 indicates instability. Resilience is defined as R = 1 − ρ(J): positive in stable regimes, approaching zero at criticality.
Let a civilisation be a network of N institutional nodes with state vectors m_i(t), and define the operational projections:
$$\Psi_i(t) = A_i(t) \cdot C_i(t), \qquad \Phi_i(t) = K_i(t) \cdot S_i(t)$$
Let system coupling be Λ(t) (aggregate Bond Strength), and define rate dispersion on the cognitive plane as:
$$D_\Psi(t) = \text{std}_i(\Delta\Psi_i(t)), \qquad \Delta\Psi_i(t) = \Psi_i(t) - \Psi_i(t-1)$$
Define the criticality index:
$$\mathcal{X}(t) = \frac{D_\Psi(t)}{\Lambda(t)}$$
Assume only: (1) Incommensurate rates — there exists persistent heterogeneity in node update rates; (2) Coupling-limited synchronisation — the ability of the network to maintain coordinated evolution is increasing in Λ and decreasing in D_Ψ; and (3) Finite buffering — capacity is bounded so that prolonged stress can degrade effective coupling.
Theorem. There exists a threshold θ such that if 𝒳(t) > θ for a sustained interval, the system enters a critical regime in which a coordination phase transition occurs with elevated probability. The transition resolves into one of four attractor classes:
Crisis is a phase-space property of the coupling–dispersion ratio, not a narrative, ideological, or agentic failure mode.
Corollary 1 (Non-Agentic Crisis). A coordination transition can occur without any node's absolute state collapsing. Observable rupture can arise purely from desynchronisation. "Bad leadership," "irrationality," or "institutional incompetence" are not necessary conditions for systemic crisis.
Corollary 2 (Coupling Primacy). Year-to-year changes in system condition are dominated by changes in coupling: corr(ΔΛ, ΔH) ≫ corr(Δx_i, ΔH) for any single-node component x_i and any monotone proxy H of system performance.
Corollary 3 (Rate-Dispersion Diagnostic). Dispersion measured on levels (e.g., σ_V) is not a reliable early-warning statistic across regimes, but dispersion measured on rates (D_Ψ) is. Early-warning must be computed from delta-fields, not state snapshots.
Corollary 4 (Bifurcation, Not Cliff). High criticality does not imply a unique direction of change. When 𝒳 > θ, the system's next move can be either forced re-coherence or fracture, depending on repair capacity and constraint geometry. Crisis indicators identify windows of phase plasticity, not inevitable collapse.
Corollary 5 (Late Abstraction Collapse). If abstraction collapse occurs, it is generically lagged behind sustained high-𝒳 and degraded Λ. Falling abstraction is a downstream impairment of prolonged coordination damage, not its cause.
Corollary 6 (Attractor Class Identifiability). A system's coordination attractor class can be identified empirically from the conditional response of coupling to criticality: 𝔼[ΔΛ(t+1) | 𝒳(t) > θ] — positive indicates re-synchronisation, near zero with alternating sign indicates oscillation, negative indicates fracture, and low 𝒳 indicates buffering.
Corollary 7 (Ontology of Approximation). If distinct datasets produced by imperfect scoring nonetheless reproduce strong ΔΛ–ΔH alignment and consistent critical regimes preceding large coupling moves, then the variables m_i are structurally real in the minimal sense required for explanation: coherent second-order dynamics cannot be generated by uncorrelated measurement noise.
CAMS uses thermodynamic language — entropy, temperature, far-from-equilibrium — throughout its theoretical apparatus. To forestall misreadings, we distinguish three epistemic layers:
Analogue layer (metaphor). "Entropy export" and "metabolic network" are structural metaphors mapping the mathematics of dissipative systems onto institutional dynamics. They do not imply that societies literally obey the second law in SI units.
Operational layer (defined variables). Entropy and temperature are treated as effective measures of uncertainty and disequilibrium appropriate to open, information-processing systems, aligned with information-theoretic entropy (Shannon 1948) and maximum-entropy inference (Jaynes 1957). Shannon entropy measures disorder/uncertainty in any system with probabilistic states, without requiring heat-bath assumptions. Jaynes' maximum-entropy principle connects information theory to statistical mechanics, providing the canonical bridge between "entropy as uncertainty" and "entropy as thermodynamic quantity."
Validation layer (tested predictions). The thermodynamic vocabulary generates predictions (stress-capacity anti-correlation, coupling-mediated crisis, late abstraction collapse) that are tested against empirical data. The vocabulary is retained because the predictions hold.
This three-layer framing is standard in the Haken/Kauffman complexity tradition. It allows reviewers to evaluate CAMS claims at each level independently.
Node state variables are scored by independent AI assessors (Gemini, Grok, GPT-4) operating under identical CAMS instruction sets. Each assessor independently generates Coherence, Capacity, Stress, Abstraction, and Bond Strength values for each node at each time step, drawing on its training corpus as a proxy for historical and contemporary knowledge.
The scoring methodology invites an immediate objection: are the scores measuring real institutional properties, or merely reflecting patterns in AI training data?
The cross-LLM concordance results (Section 6.4) provide the empirical answer. When three independently trained AI systems, with different architectures, training corpora, and known biases, converge on the same structural dynamics — the same stress trajectories, the same coupling patterns, the same crisis timing — the concordance cannot be explained by shared noise. Uncorrelated measurement error does not produce coherent second-order dynamics. The variables are structurally real in the minimal sense required for scientific explanation: they track something in the historical record that is consistent across independent observers.
This argument connects to the near-decomposability literature (Simon 1962): in systems with hierarchical modular structure, coarse-grained variables can capture inter-module dynamics even when intra-module details are lost. CAMS scores are coarse-grained approximations that capture inter-nodal coordination dynamics. Their adequacy is demonstrated by the concordance of independent assessors, not by their precision.
This does not make the scores precise. It makes them ontologically safe: adequate approximations of real coordination dynamics, sufficient to detect the phase-space signatures that the theory predicts. Independent human expert coding remains the gold standard for validation and is identified as a priority for future work.
The empirical programme encompasses 18 societies spanning approximately 2,000 years of institutional history:
| Society | Period | Society-Years | Regime Type |
|---|---|---|---|
| Rome | 10–470 CE | 87 | Imperial autocracy |
| France | 1770–2025 | 256 | Multiple regime transitions |
| Japan | 1850–2025 | 166 | Imperial → democratic |
| Brazil | 1881–2025 | 145 | Imperial → federal republic |
| Germany | 1880–2025 | 146 | Imperial → fascist → democratic |
| Norway | 1881–2025 | 145 | Constitutional monarchy |
| Sweden | 1880–2025 | 145 | Constitutional monarchy |
| South Africa | 1880–2025 | 146 | Colonial → apartheid → democracy |
| Russia | 1880–2025 | 137 | Imperial → Soviet → federal |
| Australia | 1900–2025 | 126 | Federal parliamentary democracy |
| Denmark | 1900–2025 | 126 | Constitutional monarchy |
| Iran | 1900–2025 | 126 | Imperial → theocratic republic |
| Thailand | 1900–2008 | 102 | Constitutional monarchy |
| UK | 1900–2025 | 126 | Constitutional monarchy |
| Singapore | 1930–2025 | 96 | Colonial → one-party democracy |
| China | 1975–2025 | 51 | Single-party socialist republic |
| Venezuela | 1970–2025 | 56 | Federal presidential republic |
| Ukraine | 1930–2025 | 85 | Soviet → parliamentary republic |
Total: ~2,267 society-years; ~19,141 observation rows; 3 independent LLM assessors (Gemini, Grok, GPT-4); 25 dataset-assessor combinations.
Prediction. If societies are far-from-equilibrium systems subject to thermodynamic constraints, stress and capacity should be inversely related across all societies and all regime types. High stress depletes capacity; high capacity buffers against stress.
Result. 25 out of 25 dataset-assessor combinations show statistically significant negative correlation between Stress and Capacity (p < 0.05 in all cases).
| Dataset | ρ(S, K) | p-value |
|---|---|---|
| Australia (Gemini) | −0.508 | 3.9 × 10⁻⁶⁷ |
| Australia (Grok) | −0.543 | 3.4 × 10⁻⁶⁹ |
| Iran (Gemini) | −0.776 | 2.6 × 10⁻²⁰³ |
| Iran (Grok) | −0.886 | < 10⁻³⁰⁰ |
| China (Gemini) | −0.408 | 8.0 × 10⁻¹⁸ |
| China (GPT-4) | −0.527 | 1.3 × 10⁻⁷³ |
| Germany (Gemini) | −0.778 | 1.4 × 10⁻²⁶³ |
| Norway (Gemini) | −0.672 | 1.6 × 10⁻¹⁵³ |
| Norway (Grok) | −0.713 | 1.6 × 10⁻¹⁶⁹ |
| Sweden (Gemini) | −0.584 | 2.8 × 10⁻¹⁰⁶ |
| South Africa (Gemini) | −0.504 | 3.5 × 10⁻⁷⁶ |
| South Africa (Grok) | −0.687 | 7.6 × 10⁻¹⁶³ |
| Venezuela (Gemini) | −0.869 | 4.1 × 10⁻¹³⁸ |
| Singapore (Gemini) | −0.661 | 1.2 × 10⁻⁹⁹ |
| Denmark (Gemini) | −0.559 | 6.3 × 10⁻⁸⁴ |
| Brazil (Gemini) | −0.758 | 1.2 × 10⁻²¹⁸ |
| France (Grok) | −0.792 | < 10⁻³⁰⁰ |
| UK (Gemini) | −0.683 | 2.0 × 10⁻¹³⁹ |
| Russia (Gemini) | −0.660 | 3.2 × 10⁻¹³⁸ |
| Japan (Gemini) | −0.977 | < 10⁻³⁰⁰ |
| Japan (Grok) | −0.780 | 2.4 × 10⁻¹⁹⁷ |
| Thailand (Gemini) | −0.488 | 1.3 × 10⁻⁵⁰ |
| Ukraine (Gemini) | −0.726 | 1.5 × 10⁻¹¹³ |
| Ukraine (Grok) | −0.790 | 3.1 × 10⁻¹⁰⁰ |
| Ukraine (GPT-4) | −0.245 | 1.2 × 10⁻¹² |
| Mean (all 25) | −0.663 | — |
The universality is total. No society, no assessor, no historical period violates the prediction. The stress-capacity relationship also exhibits a consistent negative slope across societies (ranging from −0.31 to −0.84), indicating that the thermodynamic constraint operates at similar intensities across radically different governance architectures.
Prediction. If Bond Strength functions as an order parameter for coordination, it should correlate strongly with mean Node Value (the proxy for aggregate system health).
Result. 24 out of 24 dataset-assessor combinations show strong positive correlation (mean ρ = +0.929). The weakest result is Japan (Grok) at ρ = +0.631; the remainder all exceed ρ = +0.78.
This is not a trivial finding. In the theoretical formulation (Section 3.3), the coupling matrix W(t) is reconstructed from nodal states — so coupling is formally a function of the state variables. But in the empirical protocol, Bond Strength is scored independently by each assessor as a separate judgement. The near-perfect correlation between independently scored Bond Strength and Node Value is a meaningful finding precisely because it is not definitional. Assessors arrive at coupling estimates separately from the state variables, yet the two track almost perfectly. That is evidence that coordination quality is a real structural quantity, not a scoring artefact.
Prediction. If the framework measures real coordination dynamics rather than artefacts of any single AI system's training bias, independently trained assessors should converge on structural trajectories while diverging on absolute levels and cultural texture.
Result. Eight cross-LLM comparison pairs:
| Pair | Common Years | Bond ρ | Stress ρ | Node Value ρ |
|---|---|---|---|---|
| Australia: Gemini vs Grok | 111 | 0.618 | 0.647 | 0.692 |
| Iran: Gemini vs Grok | 126 | — | 0.782 | 0.861 |
| Norway: Gemini vs Grok | 136 | 0.767 | 0.694 | 0.780 |
| South Africa: Gemini vs Grok | 145 | 0.666 | 0.618 | 0.727 |
| Ukraine: Gemini vs Grok | 46 | 0.508 | 0.569 | 0.615 |
| Ukraine: Gemini vs GPT-4 | 85 | 0.801 | 0.744 | 0.851 |
| China: Gemini vs GPT-4 | 51 | 0.853 | 0.752 | 0.844 |
| Japan: Gemini vs Grok | 115 | −0.130 | 0.557 | 0.477 |
Mean concordance: Bond Strength ρ = 0.583; Stress ρ = 0.670; Node Value ρ = 0.731.
The concordance pattern is itself informative. Material-constraint variables (Stress, Capacity, and their composite Node Value) show stronger cross-LLM agreement than the more interpretive coupling variable (Bond Strength). This is precisely what the ontological-safety argument predicts: LLMs converge on material constraints while diverging appropriately on interpretive quantities.
The single anomalous pair (Japan, Bond Strength ρ = −0.130) reflects known differences in how Gemini and Grok handle the Meiji-to-Shōwa transition — a period where the mapping between traditional institutional categories and CAMS nodes is genuinely contested. The Stress and Node Value concordance for Japan remains positive, confirming that material dynamics are captured even when coupling interpretation diverges.
Prediction. Flow (energy throughput) and Stewards (control over energy conversion) are the primary drivers of system health. Their node values should consistently rank among the strongest correlates of the system mean.
Result. Across societies with standard node naming, Flow appears in the top three system-health correlates in 11 of 20 datasets, and Stewards in 10 of 20. Craft also emerges as a consistent top contributor (14 of 20), reflecting its intermediate position between the fast and slow quartets. Shield and Lore consistently rank lowest.
The metabolic core finding has a direct empirical implication tested in a separate validation study: when Stewards capacity falls below Stewards stress (the "rentier buffer collapse"), Shield activation follows within 1–3 years, with a mean increase of +0.468 and 66.7% probability of escalation (N = 102 crisis periods across four societies and 255 years of data, p < 0.0001, Cohen's d = 0.710). When Lore-Archive coupling is weak, the kinetic export effect is 520% greater than when symbolic recovery capacity is intact.
This confirms that military expansion is a downstream consequence of metabolic stress — an attempt at entropy externalisation that routinely fails due to second-order coupling effects. War is a symptom of coordination failure, not a cure for it.
Prediction. If coordination physics is universal, societies classified as geopolitical rivals should exhibit similar node-ranking profiles.
Result. Spearman rank correlations between societies' mean node-value profiles:
| Pair | ρ | Interpretation |
|---|---|---|
| China vs Russia | +0.833 | Very similar |
| China vs Iran | +0.786 | Very similar |
| Iran vs Russia | +0.714 | Very similar |
| Germany vs Sweden | +0.881 | Very similar |
| Denmark vs Norway | +0.707 | Very similar |
| Denmark vs Sweden | +0.707 | Very similar |
| Brazil vs Russia | +0.881 | Very similar |
| Iran vs South Africa | +0.810 | Very similar |
| Australia vs Brazil | +0.810 | Very similar |
| Singapore vs South Africa | +0.738 | Very similar |
Societies sharing geographic-demographic constraints show the strongest structural similarity, regardless of governance type. China (single-party socialist republic), Russia (federal authoritarian), and Iran (theocratic republic) share more coordination structure with each other — and with Western democracies — than political analysis would predict.
The finding that "rival" societies face similar coordination physics is not a political claim. It is a structural observation: the constraints that shape civilisational coordination are thermodynamic and geographic, not ideological.
The theoretical development of CAMS has been iterative, but the direction of iteration is significant. The framework was not adjusted to produce interesting results. The data revealed dynamics so clear and so universal that successive theoretical refinements were required to account for what was being observed.
Three findings were particularly unexpected:
The universality of stress-capacity anti-correlation. That all 25 dataset-assessor combinations would show the same directional relationship, across regime types from Imperial Rome to contemporary Singapore, was not guaranteed by the theory. The theory predicts it; the data confirms it without exception.
The consistency of Shield's low predictive power. Shield (security/defence) consistently ranks among the weakest contributors to system health across diverse societies. This directly contradicts realist geopolitical assumptions that security capacity drives civilisational vitality. In the CAMS framework, Shield functions as a heat-dissipation mechanism — it activates when metabolic stress (primarily in Flow and Stewards) cannot be processed internally. Security expansion is a symptom of coordination failure, not a cause.
The convergence of independent AI assessors. Three LLMs with different architectures and training corpora, given identical analytical instructions, converge on the same structural dynamics. This convergence is evidence that the dynamics are in the historical record, not in the models.
The Coordination Phase Transition Theorem's most consequential implication is that crisis does not require villains. A society can enter a critical regime — and undergo a genuine coordination phase transition — without any individual node collapsing, any leader failing, or any ideology prevailing. The crisis is structural: it arises from the desynchronisation of institutional timescales when coupling capacity is insufficient to absorb rate dispersion.
This reframing has direct analytical consequences. If crisis is a phase-space property rather than a narrative of failure, then the appropriate response is architectural (strengthen coupling, reduce rate dispersion) rather than punitive (identify and replace failing agents). Policy interventions targeting Flow and Stewards — the metabolic core — should yield substantially greater leverage on system health than equivalent investments in Shield, a prediction with clear empirical implications.
The constraint-similarity findings challenge the empirical basis of competitive geopolitical framings. If China, Russia, and Iran face the same coordination physics as Western democracies — the same stress-capacity tradeoffs, the same coupling-mediated crisis dynamics, the same metabolic core vulnerabilities — then characterising these societies as existential threats on the basis of ideological difference is structurally unfounded.
This does not imply that all societies are equivalent, or that geopolitical tensions are illusory. It implies that the constraints are shared, even when the responses differ. All complex societies face the same thermodynamic limits. The path toward mutual stability is not dominance but shared architectural reform: strengthen internal recovery mechanisms, decouple rentier-state synchronisation, rebuild symbolic legitimation capacity.
Scoring circularity. The reliance on AI-generated scores creates a methodological loop. The cross-LLM concordance results mitigate but do not eliminate this concern. Independent human coding validation — historical experts scoring node states without access to AI outputs — is the gold standard and has not yet been completed.
Dimensional optimality. The rate-separation argument for eight nodes is formalised as a model-order selection problem (Section 2.4) but the comparative model fitting (N ∈ {5,6,...,12}) has not been performed. Eigenmode analysis demonstrating that eight principal components capture >90% of variance in the 32-dimensional state space would substantially strengthen the architectural claim.
Prospective validation. The empirical results are retrodictive. Genuinely prospective testing — sealed predictions against future outcomes — is planned for 2026–2028.
No peer review. This work has not undergone formal external review.
The eight-node architecture of CAMS is not a classification system. It is a minimal coordination model derived from the stability requirements of multi-timescale complex adaptive systems. The rate-separation argument establishes why eight — and not five, ten, or twenty — functional nodes are required for tractable description of civilisational coordination dynamics. The dimensional optimality claim is formalised as a model-order selection problem and is empirically adjudicable.
The mathematical framework connects CAMS to the Kuramoto synchronisation literature and the Master Stability Function formalism. The criticality index 𝒳(t) = D_Ψ/Λ is structurally identical to the disorder-to-coupling ratio governing phase transitions in coupled oscillator networks. Algebraic connectivity of the coordination Laplacian provides a computable, falsifiable diagnostic for synchronisation capacity. This is not analogy — it is the same mathematics operating on institutional substrates.
The empirical record, tested across 18 societies, approximately 2,267 society-years, and three independent AI assessors, reveals coordination dynamics of striking universality: stress and capacity are universally anti-correlated; bond strength is a near-perfect tracker of system health; metabolic core nodes (Flow and Stewards) dominate system dynamics while security apparatus (Shield) consistently ranks lowest; and independent assessors converge on the same structural trajectories.
Crisis, in this framework, is not a failure of will, intelligence, or ideology. It is a coordination phase transition — a structural desynchronisation that arises when rate dispersion overwhelms coupling capacity. The outcome depends on the system's attractor-class repair response, not on the moral quality of its leadership.
The clarity of the dynamics observed in the data has driven the theoretical development. CAMS is the current model adequate to that signal. Whether it achieves disciplinary acceptance depends on surviving prospective validation and peer review. The framework has, at minimum, earned the right to be tested seriously.
| CAMS Component | Scientific Foundation | Key Reference |
|---|---|---|
| Rate separation | Multi-timescale dynamical systems | Haken (1983) |
| Coupling network | Graph synchronisation, coupled oscillators | Kuramoto (1984); Pecora & Carroll (1998) |
| Dimensional optimality | Model-order selection, MDL/AIC | Rissanen (1978); Akaike (1974) |
| Near-decomposability | Hierarchical modular systems | Simon (1962) |
| Information-theoretic entropy | Shannon entropy, MaxEnt inference | Shannon (1948); Jaynes (1957) |
| Early-warning signals | Critical slowing down | Scheffer et al. (2009) |
| Institutional differentiation | Functional differentiation in social systems | Luhmann (1995) |
| Causal architecture | Information flow, Granger causality | Barnett & Seth (2014) |
| Psychosocial operators | Constructed cognition, affective systems | Barrett (2017) |
| Historical dynamics | Cliodynamics, secular cycles | Turchin (2003) |
| Debt-kinetic dynamics | Compound interest divergence, rentier capture | Hudson (2018); Piketty (2014) |
Akaike, H. (1974). "A new look at the statistical model identification." IEEE Transactions on Automatic Control, 19(6), 716–723.
Bar-Yam, Y. (2003). Dynamics of Complex Systems. Westview Press.
Barnett, L. & Seth, A.K. (2014). "The MVGC multivariate Granger causality toolbox." Journal of Neuroscience Methods, 223, 50–68.
Barrett, L.F. (2017). How Emotions Are Made. Houghton Mifflin Harcourt.
Haken, H. (1983). Synergetics: An Introduction. Springer-Verlag.
Holland, J.H. (1995). Hidden Order: How Adaptation Builds Complexity. Addison-Wesley.
Hudson, M. (2018). ...and forgive them their debts. ISLET-Verlag.
Jaynes, E.T. (1957). "Information theory and statistical mechanics." Physical Review, 106(4), 620–630.
Kauffman, S. (1993). The Origins of Order. Oxford University Press.
Kuramoto, Y. (1984). Chemical Oscillations, Waves, and Turbulence. Springer-Verlag.
Luhmann, N. (1995). Social Systems. Stanford University Press.
McKern, K.F. (2025–2026). Complex Adaptive Humans newsletter series. LinkedIn.
Pecora, L.M. & Carroll, T.L. (1998). "Master stability functions for synchronized coupled systems." Physical Review Letters, 80(10), 2109–2112.
Piketty, T. (2014). Capital in the Twenty-First Century. Harvard University Press.
Rissanen, J. (1978). "Modeling by shortest data description." Automatica, 14(5), 465–471.
Scheffer, M. et al. (2009). "Early-warning signals for critical transitions." Nature, 461, 53–59.
Shannon, C.E. (1948). "A mathematical theory of communication." Bell System Technical Journal, 27(3), 379–423.
Simon, H.A. (1962). "The architecture of complexity." Proceedings of the American Philosophical Society, 106(6), 467–482.
Turchin, P. (2003). Historical Dynamics: Why States Rise and Fall. Princeton University Press.
West, G.B. (2017). Scale. Penguin Press.
Correspondence: Kari Freyr McKern, Complex Adaptive Humans Data and analysis code available on request.