The CAMS Model

Complex Adaptive Model of Societies — v1.0-Final · locked 7 June 2026

CAMS (Complex Adaptive Model State) is a graph-theoretic diagnostic framework for tracking the internal coordination health of societies as complex adaptive systems. It is a diagnostic instrument, not a predictive oracle: common-mode signal detects that a system is under stress; the residual (node-level) layer identifies which function is failing.

The v1.0-Final formulation (locked 7 June 2026) resolved the three architecture decisions that had been open since the April 2026 release candidate: node architecture, the structural adjacency prior, and the canonical edge-weight equation. All cross-society comparison must use values recomputed under this formulation from raw scores — pre-existing Node Value or Bond Strength columns in older datasets should never be treated as already compliant.

CAMS models any persistent organised collective through eight invariant functional nodes, each corresponding to a coordination requirement every society must meet in some form.

NodeFunctional roleTypical institutional referents
HelmExecutive / strategic leadershipHead of state, cabinet, ruling party, legitimation
ShieldDefence / coercion monopolyMilitary, police, territorial security
LoreCultural narrativeReligion, ideology, media, education
StewardsBureaucratic administrationCivil service, regulators, fiscal apparatus
ArchiveKnowledge & institutional memoryLibraries, universities, records, legal codes
CraftProduction & manufacturingIndustry, artisanal economy, resource extraction
HandsLabour & collective actionWorkforce organisation, labour markets, guilds
FlowDistribution & marketsTrade networks, transport, currency, supply chains

Architecture decision (locked): The eight-node partition is retained. A proposed Lore–Archive merger was formally rejected — the pair shows the strongest residual co-movement in the corpus (mean r = 0.643 after removing common-mode), but this reflects shared exposure to slow-loop stressors, not functional identity. Archive is institutional memory; Lore is the normative/cultural layer. A separately mooted Craft–Flow merger has been withdrawn outright.

Each node is scored on four dimensions, integer 1–10: Coherence (C) — internal consistency; Capacity (K) — institutional depth and output; Stress (S) — external pressure and adaptive demand; Abstraction (A) — institutional sophistication and symbolic reach.

Node Viability

V_i = C_i + K_i − S_i + 0.5·A_i

Cognitive Activation (singularity-only floor)

σ_i = (A_i · C_i / 100) · (K_i − S_i)

The Ki = Si singularity is the only point where a floor substitution (0.1) is applied — not as a blanket clip across all values. An earlier draft that floor-clipped every node made σmin structurally non-negative and silently disabled the Local Node Failure trigger; that form is retired.

Coupling Quality

q_i = (0.6·C_i + 0.4·A_i) / 10

Bounded Edge Weight

B_ij = T_ij · √(q_i · q_j) · 2^(−(S_i + S_j)/10)

Structural adjacency prior (locked): Tij = 1 for all node pairs — a fully connected graph. A theoretically motivated sparse prior was tested and shifts magnitudes but not diagnostic rank ordering at current scoring granularity; a data-driven Tij is deferred to v1.1, once the corpus is large enough to estimate pairwise bond trajectories empirically.

Connectivity

λ₂ — second eigenvalue of the raw, unnormalised Laplacian (observed range ≈ 0.3–3.4)

The normalised Laplacian collapses to a degenerate [1.07–1.14] range across societies and must not be used for cross-society comparison.

Φ_G(t) = (V̄, σ_V, V_min, B̄, λ₂, σ_min)

This replaces an earlier two-dimensional classifier built on (V̄, σ_V) alone. The two added terms — Vmin and σmin — exist specifically to catch single-node failures that the two-dimensional version missed (Germany 2024's Helm collapse, USA 2020), where the system mean looked unremarkable while one node had already failed.

Local Node Failure trigger

V_min < 4.0 OR σ_min ≤ −0.85 — independent of V̄

This is the single most consequential threshold in the classifier, because it is the only one that doesn't require the whole system to be visibly in trouble before flagging a failing part.

Regime thresholds beyond this trigger are provisional and corpus-calibrated on a small sample (8 societies); recalibration across the full ~50-series corpus is an open item, not yet part of the lock.

CAMS is a graph-theoretic diagnostic model. Eight nodes collapse to roughly 2–3 effective statistical dimensions on the current corpus (PC1 alone explains 47–89% of variance depending on society) — the four scoring dimensions are best understood as theoretical lenses on a smaller number of empirical degrees of freedom, not four independent measurements. Thermodynamic and Kuramoto-style language used elsewhere in the project is analogical scaffolding, not a physically derived claim.

Where V̄ is very low (≲ 2), the whole-system common-mode signal dominates and residual node attribution becomes unreliable — this caveat applies wherever a crisis is attributed to a specific node.

All societies in the corpus — Western and non-Western, ancient and modern, democratic and authoritarian — are scored by the same operators without geopolitical privilege. Where the model's output touches on threat discourse (e.g. external-state framing), the structural metrics, not an essentialised account of the external actor, are what's being reported.

v1.0-Final — operators locked 7 June 2026 Not yet validated out-of-sample

v1.0-Final is closed-form and locked for the operators above. It is not yet validated out-of-sample: that is the job of the 2026–2028 prospective prediction set, pre-registered with explicit falsification criteria, which is the primary pathway toward a stronger framework rating. Until those predictions resolve, claims should be framed as "strongly supports" rather than "demonstrates."

CAMS v2.3 — Stable canonical framework v3.2-R — Research-grade operator extension (experimental)

The v3.2-R extension adds ESCH σ (entropy), κ (capacity fraction), headroom, and attractor-state operators used in CAMS Explorer and CAMS Interpreter. It is an experimental research tool, not a new framework version. The canonical framework remains v2.3 as tagged in the README and research diary.

Neural Nations applies the Complex Adaptive Model of Societies (CAMS) — a physics-inspired, scale-covariant framework that treats societies (civilisations, nations, companies, departments) as coupled institutional networks rather than narrative-driven stories.

Instead of ideology or single-cause explanations, CAMS models any social system as an 8-node × 4-metric matrix evolving over time. It captures coordination failures, stress accumulation, phase transitions, and resilience — before collapse becomes visible.

Core idea in one sentence: Societies fail not from one bad node collapsing, but from severed bonds between Mythic (meaning-making), Interface (executive/coordinating), and Material (productive) layers — measurable as declining cross-layer coupling Λ(t).

Universal functional "organs" that appear in every stable society, empire, or organisation — regardless of culture, era, or scale.

1. Lore Mythic core: shared stories, ideology, cultural memory that binds identity.
2. Archive State/knowledge memory: records, institutions preserving continuity and legitimacy.
3. Helm Executive/strategic centre: decision-making, policy direction, "brain" of the system.
4. Stewards Elite/property owners: resource controllers, guardians of capital and hierarchy.
5. Shield Military/security/defence: protective boundary, coercion capacity.
6. Craft Knowledge workers/professions: specialists, innovators, technical elite.
7. Hands Labour/proletariat: productive base, mass execution force.
8. Flow Merchants/trade/economy: circulation, exchange, resource distribution networks.

Each node is evaluated blindly across four orthogonal dimensions.

Coherence (C) Internal alignment and consistency within the node.
Capacity (K) Resources, capability, and effectiveness.
Stress (S) Accumulated pressure, entropy, dysfunction.
Abstraction (A) Symbolic sophistication and long-range planning capacity.

Node value at time t:

Vᵢ(t) = Cᵢ + Kᵢ + (Aᵢ / 2) − Sᵢ   range ≈ [−7.5, 24.0]
Mean societal value: V̄(t) = (1/8) Σ Vᵢ(t) Dispersion (tension): σ_V(t) = √[ (1/8) Σ (Vᵢ − V̄)² ] Bond strength i↔j: Bᵢⱼ(t) = √[ max(Vᵢ+8, 0) × max(Vⱼ+8, 0) ] / 32 ∈ [0, 1] Cross-layer coupling: Λ(t) = mean Bᵢⱼ over cross-layer edges

Full specification: bond-strength-spec.html

Dynamics update (diffusion + noise, analogous to Ising/spin-glass coordination models):

Vᵢ(t+1) = Vᵢ(t) + α Σⱼ Bᵢⱼ (Vⱼ − Vᵢ) + εᵢ(t+1)

Coordination phase space: Φ(t) = (V̄(t), σ_V(t))

Regime V̄(t) σ_V(t) Λ(t)
Coherent-Capable > ~12 < ~3.5 High
Crisis / Transition Low High Falling
Coordination failure Λ(t*) < 0.45 — discharge via Shield or collapse < 0.45

CAMS emerged from Occam's razor: after testing neural-net approaches, the simplest structure (8 functions × 4 metrics) proved cross-culturally robust and predictive (ensemble r > 0.7 across LLMs, 5+ year lead times on historical benchmarks including 1861 USA onset).

It maps directly to positive/negative feedback loops, hysteresis in institutional change, conductivity (shared abstraction enabling coordination), and thermodynamic-like entropy flows in stressed systems.

sybond_report_exporter.py — Generates a fully-structured 10-section Sybond Report in Markdown from a CAMS ensemble mean CSV. Supports blank templates and data-filled reports using the v3.2-R working kernel operators (Vi, σi, V̄, σV, Library Attractor proxy ηloop). Accepts flexible column aliases; handles both mean and uncertainty-envelope CSVs.

# Blank template python sybond_report_exporter.py --blank --society France --sybond-name Marianne --output France_template.md # Data-filled report python sybond_report_exporter.py \ --mean-csv France_CAMS_ensemble_mean.csv \ --envelope-csv France_CAMS_envelope.csv \ --society France --sybond-name Marianne \ --output France_Sybond_Report.md

Requires pandas. Column names are resolved via flexible aliases (e.g. coherence / c / c_i all map to C). Node names are canonicalised automatically from common synonyms.

View on GitHub

Everything — raw formulas, Python implementation (cams_framework_v2_1.py, cams_engine.py), validation reports, reproducibility notes — lives in the open repository.

GitHub: KaliBond/wintermute

See README.md, CAMS_INDEX.md, CAMS_Validation_Formulation.md, and DATASET_VALIDATION_SUMMARY.md for complete specs.

← Research Diary Public Datasets → Explore Artefacts → Open Dashboard →