FAQ

What does CAMS stand for, and what does it actually do?

CAMS stands for Complex Adaptive Model State. It models any society — or any sufficiently complex organisation — as a network of eight functional nodes scored across four metrics. The output is a set of time series showing how each node has evolved, how strongly the nodes are coupled, and where the system sits in the space between stable coordination and coordination failure.

The eight nodes are: Helm (executive governance), Shield (military and defence), Lore (cultural memory and ideology), Archive (institutional memory and records), Stewards (property and elite resource control), Craft (knowledge workers and technical capacity), Hands (labour), and Flow (trade and economic circulation).

What problem is Neural Nations trying to solve?

Most frameworks for understanding societies are either purely qualitative — historical narrative, political theory — or they focus on specific quantitative proxies like GDP, Gini coefficient, or democracy indices that don't capture how a system is coordinating as a whole. CAMS tries to occupy the gap: a structured, measurable, cross-culturally comparable instrument for institutional health.

The practical question it asks is: can you tell, from structure and trajectory alone — not from ideology, not from outcome — which direction a society is moving and how much stress it is absorbing?

Why eight nodes? Why not more or fewer?

CAMS started with a much larger set of candidate functions and was progressively reduced. Eight turned out to be the minimum number of functions that is (a) universally present across all stable societies regardless of era, culture, or scale, and (b) sufficient to detect the structural patterns that precede major historical transitions.

The argument for this minimal architecture is written up formally in the draft paper "Why Eight Nodes?" in the GitHub repository.

Is CAMS peer-reviewed?

No. CAMS has not yet been formally peer-reviewed. This is stated clearly because it matters. The work is published openly — datasets, formulas, scoring protocols, validation reports — precisely so that scrutiny is possible.

Scholarly critique and independent replication are the explicit goal, not an afterthought. The Validation & Limits page states exactly what has been demonstrated and what has not. The Contact page has a specific track for peer review and co-authorship enquiries.

Are the equations derived from first principles or calibrated empirically?

The Kepler analogy applies: Kepler mapped planetary orbits with extraordinary precision before Newton explained why they moved that way. CAMS is in the same position — it fits observed societal behaviour with consistent retrodictive accuracy, but it does not yet have the formal derivation that a mature scientific theory requires. The formal derivation is ongoing work.

This is an honest statement of where the research programme currently stands, not a limitation to be embarrassed about.

How are scores produced?

Each node is scored on a 1–10 scale across four dimensions — Coherence, Capacity, Stress, and Abstraction — for each year in the dataset. Scoring is done by multiple AI language models working independently from the historical record. Depending on the series, between three and five scorers are used.

Results from multiple scorers are compared for concordance. High concordance across independently trained systems suggests the signal is in the historical record rather than in any one scorer's assumptions.

What does "blinded" scoring mean in practice?

The blinding protocol means that when a scorer evaluates, say, the United States in 1925, it is given only the historical information available at that time — not the outcome of the Depression, the New Deal, or World War Two. The scorer is asked to assess the node's function in that year without knowing what happened next.

This matters because the standard critique of any retrospective model is that it was calibrated to fit known outcomes. Blinding makes that critique harder to sustain, though it doesn't eliminate it: the scorers are trained on historical text that may implicitly contain outcome information. This limitation is acknowledged openly in the methodology documentation.

What does r > 0.7 mean here?

r is the Pearson correlation coefficient — a measure of how consistently two variables move together, ranging from −1 (perfectly inverse) to +1 (perfectly aligned). In CAMS, "ensemble r > 0.7" means that when multiple independent AI scorers score the same society for the same year, their outputs correlate above 0.7: they are largely agreeing on the structural trajectory even though they are working independently.

This concordance threshold is used to validate that the signal is consistent across scorers, not just an artefact of one model's assumptions or training data.

How should I interpret "prediction", "hindcast", and "validation" claims?

• Hindcast means applying the model to known historical data and testing whether the model would have detected the right structural signals. Ensemble hindcast accuracy of 75–90% across tested societies is the strongest evidence CAMS currently has.
• Validation means cross-checking CAMS outputs against independent datasets — for example, the Seshat Databank cross-validation, which yielded r = 0.78 on Latium/Rome — or testing concordance across multiple independent scorers.
• Prospective prediction — applying the model to the present and testing whether it is right about the future — is the test CAMS has not yet had time to pass or fail. It is the explicitly stated next step in the research programme.

How large is the dataset and what does it cover?

As of April 2026: 45 historical series, 38 distinct societies and organisations, 39,351 node-year records. The time span runs from 5 CE to 2026, though coverage is densest from 1800 onward. Societies include nation-states, empires, city-states, and corporate organisations. All datasets are freely downloadable from the Datasets page and from GitHub.

What is the difference between CAMS, CAMS v2.3, CAMS v2.3.1, and CAMS v3.2-R?

• CAMS refers to the framework as a whole — the name of the measurement system.
• CAMS v2.3 is the current stable canonical specification — the version documented on the Model page and tagged in the GitHub README. This is the version all datasets use.
• CAMS v2.3.1 is a conditional revision from April 2026 to the Compression Theorem. It clarifies that the eight-node partition is regime-dependent rather than a universal upper bound. It does not change the scoring formulas or the dataset schema.
• CAMS v3.2-R is an experimental research extension, not a new framework version. It adds operator notation — ESCH σ, κ, headroom, attractor states — used in the Explorer and Interpreter tools. Think of it as a specialist instrument built on top of the stable v2.3 foundation.

If you see a version label in the Research Diary that doesn't appear in the model changelog — such as "CAMS PRIME v3.0" (a working label from the January 2026 Sweden analysis) — treat it as a historical working name that has been superseded by the canonical v2.3 specification.

Which version should I use for my own analysis?

CAMS v2.3 — the stable canonical specification documented on the Model page. The datasets and scoring protocols published in the GitHub repository all follow v2.3. Unless you are specifically working with the Explorer or Interpreter tools and need the v3.2-R operator extensions, v2.3 is the right starting point.

What can CAMS do well?

• Detect accumulating structural stress in institutional networks before crisis becomes visible in conventional indicators.
• Compare societies across radically different cultures, eras, and scales using a common measurement framework.
• Distinguish between societies that absorb stress and adapt versus those moving toward coordination failure.
• Identify which specific institutional nodes are under the most pressure in a given period.
• Provide a structural rather than ideological lens for geopolitical and historical analysis — one that does not privilege any particular political system by design.

What can CAMS not do?

• Reliably predict specific events, dates, or outcomes. CAMS identifies trajectory and structural risk — not event timing.
• Provide exact measurements. Scores should be read as ordinal risk bands, not precise quantities.
• Capture sudden-onset fractures that bypass gradual desynchronisation. CAMS is better at detecting accumulating stress than rapid rupture.
• Replace judgement. The model is an instrument for making patterns visible, not a substitute for interpretation.
• Claim the status of settled science. It is a research programme: structured, empirical, and falsifiable — but not yet independently replicated at scale by external human research teams.

What would strengthen or weaken CAMS as a framework?

The Validation & Limits page addresses this in full. In brief:

Would strengthen: blinded prospective case testing (2026–2028), independent human replication of scoring, comparison with simpler baseline models, publication of failed and ambiguous cases.

Would weaken: repeatedly missing known crises, flagging stable systems as high risk, performing no better than simpler indicators, heavy dependence on a single scorer or dataset.

Why does the framework apply to corporations as well as nations?

CAMS is scale-covariant: the same eight functional problems — executive leadership, defence and security, cultural memory, institutional records, elite resource control, knowledge capacity, labour, and circulation — appear in every stable complex organisation, not just nation-states. A corporation, a government department, a medieval city-state, and a contemporary nation-state all have to solve these problems or cease to function.

The Boeing analysis (1990–2025) and the Qantas, Nexperia, and SpaceX datasets in the open data repository demonstrate the corporate application. The node mapping differs slightly — for example, "Shield" maps to security and legal capacity in a corporate context — but the measurement logic is identical.

Can I download and reuse the datasets?

Yes, freely. All 39,351 records across 38 societies and 45 historical series are available for download on the Datasets page and from the GitHub repository at KaliBond/wintermute. The repository is public, AI scraping is explicitly permitted, and forks and contributions are welcome. No login is required for anything on this site.

How often are datasets and tools updated?

The dataset is actively extended. Each significant addition is documented in the Research Diary. The Streamlit dashboard updates automatically when new data or features are pushed to GitHub.

Why are some materials hosted on external platforms?

The live dashboard runs on Streamlit Cloud because interactive computation at scale requires a live server. Some artefacts in the Explore section were produced using external tools during the research process. Canonical outputs — datasets, formulas, scoring protocols, and validation reports — are hosted in the GitHub repository or on neuralnations.org directly.

Does the site collect personal data?

No personal data is collected or stored when you browse neuralnations.org. The dashboard carries the same statement. Google Analytics is present on the site for aggregate traffic measurement — no personally identifiable information is retained. No cookies requiring consent are set beyond what Analytics uses, and no advertising or third-party tracking is present.

Where should a complete beginner start?

The Research Diary. Pick any entry that sounds interesting. Individual entries are written to be readable independently — no prior knowledge of complexity science or the full framework is required. The Start Here page also has a plain-language sixty-second explainer if you want context before diving into a specific entry.

How can I get in touch — as a researcher, practitioner, journalist, or curious reader?

The Contact page has three structured tracks: Scholar & Peer Review (for academics and methodologists), Practitioner & Policy (for analysts and strategists), and General Public & Press (for journalists, educators, and curious readers). Each track explains what kind of engagement is possible and what information is most useful to include when writing.

Is the companion book available?

The Architecture of Civilisation: A Thermodynamic Science of Societal Dynamics (First Edition) is complete and in preparation for release. Updates will appear in the Research Diary and the Complex Adaptive Humans LinkedIn newsletter.