CAMS Suite · v3.2-R The technical guide.
A working reference for the four CAMS apps in this project — what each metric means, where it comes from, how to read the screens, and which questions the framework can and cannot answer.
1Overview
What CAMS measures, in one paragraph.
CAMS — the Complex Adaptive Model State framework — represents a society as a network of eight institutional nodes, each scored on four measures at every observed year. The eight nodes are organised into three coupled layers — Mythic (meaning), Interface (governance), Material (production) — and the cross-layer coupling Λ(t) is the load-bearing systemic indicator.
Once a society's matrix M[node][measure] is loaded for a given year, a deterministic compute kernel derives ~25 scalar fields and one weighted graph. This guide describes those fields, the four apps that visualise them, and how to read what the apps display.
matrix[8 nodes][4 measures] → compute(matrix) → { V, B, σ, λ₂, R, κ, η_loop, headroom, attractor, alarmConfidence, … }
2The eight institutional nodes
Three layers · eight nodes · stable across all societies and time periods.
Every CAMS dataset uses the same eight nodes regardless of era or culture. The names are chosen to be transposable — Helm, Lore, Hands — rather than tied to modern institutional labels (Cabinet, University, Workforce). This is what makes Rome 200 CE comparable with Argentina 1976.
3The four measures
Each node scored on four orthogonal dimensions. Range 1–10, ensemble-averaged across raters.
| Measure | Symbol | What it captures | Range | |
|---|---|---|---|---|
| Coherence | C | Internal alignment of the node — does it know what it is for? | 1–10 | |
| Capacity | K | Resources, headcount, and effectiveness available to the node. | 1–10 | |
| Stress | S | Accumulated pressure, dysfunction, unmet demand on the node. | 1–10 | |
| Abstraction | A | Symbolic sophistication, long-range planning capacity. | 1–10 |
Measures are scored independently per rater (currently a 3- or 5-AI ensemble — GPT, Grok, Gemini, with Claude added in some panels) and averaged. Inter-rater standard deviation is published alongside each ensemble in the envelope CSV files.
4Derived metrics
From the 8×4 input matrix the kernel produces three families of derived fields.
Per-node fields
Vi is what node radius renders in the Network view. Positive V means the node is currently a contributor; negative V means it is a net liability. σi measures how much narrative force the node is applying — a low-capacity, high-abstraction node (think: a regime broadcasting myth while the lights go out) shows σ < 0.
System-wide fields
λ₂ is the textbook measure of how well a network can synchronise. Falling λ₂ signals fragmentation. R climbs when the network becomes hard to coordinate (one stiff mode, many weak ones).
Criticality & attractor signals
The κ value is binned into tiers (calibrated v3.2-R):
NOMINALκ < 0.30 WATCH0.30 ≤ κ < 0.35 WARNING0.35 ≤ κ < 0.42 CRITICAL0.42 ≤ κ < 0.57 EXTREMEκ ≥ 0.57
5Basin classification
The composite logic that names what the system is currently doing.
The kernel returns a single attractor label per year, derived from V̄, σV, κ, and Helm's V:
| Attractor | Trigger | Reading |
|---|---|---|
| Re-synchronisation | κ < 0.35 ∧ helm V ≥ 6 ∧ σV ≤ 4.5 | System is coordinating productively. Not growth, not crisis — working order. |
| Oscillation | κ ≥ 0.35 ∧ helm V ≥ 6 | Coordination is straining; cycles between loose and tight regimes. |
| Buffering | V̄ > 15 ∧ σV < 2.0 | Ample slack, low variance. Often pre-disregard — the system is not currently being challenged. |
| Fracture | helm V < 6 or σV > 4.5 or (κ ≥ 0.35 ∧ helm V < 6) | Executive decoupling, rising heterogeneity. Pre-collapse signature. |
| Thermodynamic Freeze | V̄ < 0 | Aggregate node value negative. Most nodes are net liabilities. |
Alongside the attractor, an alarm confidence level is emitted: HIGH when κ ≥ 0.35 and helm V < 6, MEDIUM when κ ≥ 0.35 alone, LOW when λ₂ < 0.5, NOMINAL otherwise.
6The four apps
Each app reads the same compute output and surfaces a different cross-section of it.
CAMS Networklive coordination view
Eight nodes, drawn at force-directed positions, with edge thickness = pairwise bond strength Wij, node radius = |Vi|, and a halo whose radius encodes σi (positive: outer ring; negative: inner ring).
How to read it
- Big nodes = high node value V — net contributors. Small nodes = drained.
- Thick edges = the two nodes are coordinated (similar C/K/S/A profile). Thin or missing = fragmented.
- Outer halo = σ > 0 (node is doing work). Inner halo = σ < 0 (broadcasting myth without backing).
- Stat strip top-of-screen: B̄, λ₂, κ, V̄, σV, current attractor.
Controls
| Society | Sidebar list — switch the dataset. |
| Year slider | Scrub through the available years; scrubbing is tween-eased so transitions are continuous. |
| ▶ / ⏸ | Autoplay through the timeline at the chosen speed. |
| Speed | Step interval (ms). Tween length tracks this. |
| Hover node | Reveals C/K/S/A and Bi for that node. |
CAMS Telescopehistorical dossier
A typewriter-styled briefing terminal for one society at one moment. Surfaces the nominal vs. real power configuration: who is formally in charge versus which node is currently doing the systemic work (highest V).
Use this when
- You want a single-page assessment of a country at a specific year — for a paper, a slide, or a sanity check.
- You want to see if Helm is still the dominant node, or whether Stewards / Shield / Lore have eclipsed it.
CAMS Interpretersignal-to-narrative bridge
For each derived signal that crosses a threshold, the Interpreter emits a signal card — a plain-language description of what to look for in the historical record, plus a Trove search string that would falsify or confirm it. This is the framework's bridge to historiography.
Each card contains
- Title — the metric & threshold that fired.
- Body — what civilisational signature this typically corresponds to.
- Confidence — HIGH / MED / LOW based on how robust the threshold is.
- Exemplar — historical years when this signature was observed.
- Trove query — a literal search string to plug into the National Library's Trove (Australian newspapers 1803–present), forming the falsification path.
Epiphenomenon Detectorsurface-vs-substrate diff
A detector for moments when the visible politics of a society (the noisy surface signal) decouples from its structural state (the slow-moving CAMS substrate). The detector flags years where what people are arguing about and what is actually changing in the institution graph have diverged.
Reading the output
- Green band: surface and structure track each other — politics is doing the work it claims to.
- Amber band: divergence growing — political theatre running ahead of (or behind) the structural reality.
- Red band: full epiphenomenal regime — surface is uncoupled, structure is changing under cover.
7Datasets
10 societies · 1800–2026 · ensemble-mean scoring across multi-AI panels.
| Society | Range | Years | Source file |
|---|---|---|---|
| USA | 1900–2026 | 127 | MARKER_USA_1900_2026_ENSEMBLE_MEAN |
| UK | 1880–2026 | 147 | UK_CAMNATIONS5_ensemble_mean |
| Germany | 1880–2026 | 147 | germany_cams5_ensemble_mean |
| Russia | 1800–2026 | 227 | russia_cams5_ensemble_mean |
| Australia | 1875–2026 | 152 | Australia_CAMS5_ensemble_mean_1875_2026 |
| Canada | 1850–2026 | 33 | Canada_CAMS5_ensemble_mean_1850_2026 |
| Argentina | 1950–2026 | 77 | Argentina_CAMS5_ensemble_mean_scores |
| China | 1850–2026 | 19 | China_CAMS5_ensemble_1850_2026 |
| Thailand | 1850–2025 | 36 | Thailand_CAMS5_ensemble_1850_2026 |
| Sweden | 1850–2026 | 19 | sweden_CAMS5_calc_1880_2026 |
Note the temporal density variance: Russia and the UK are dense annual series; Canada, Sweden and China are sparse (typically 5–10-year intervals). The compute kernel is point-wise — sparser series will appear to "tween" further during playback.
Scoring methodology
Each (society, year, node) triple is scored independently by 3–5 large language models from a blinded prompt — scorers do not know the terminal state of the society. The four measures (C, K, S, A) are returned per node, on a 1–10 scale. The published ensemble mean is the simple arithmetic mean across raters; the envelope is the per-cell standard deviation. Reliability passes the r > 0.7 threshold across raters per the Wintermute repo's validation summary.
8Recommended workflow
A canonical investigation path through the four apps.
- Telescope first. Pick a society, pick a year — get a single dossier page. This anchors your intuition before you look at signals.
- Network second. Scrub the full timeline. Look for where the graph reorganises — node sizes flipping, edges thickening or vanishing, the halo polarity changing on Helm or Lore. These reorganisations are the events.
- Interpreter third. Move to the years the Network flagged. Read which signals are firing. Note the Trove queries — you now have falsifiable predictions.
- Detector last. Ask whether the signals you saw correspond to a genuine structural shift or to surface-only political theatre. The detector tells you which.
9FAQ
Questions that come up repeatedly.
Is this just a fancy index? What stops κ from drifting wherever you want it to?
κ is a ratio of two structural quantities — the average pairwise coupling B(t) and the dispersion of the cognitive plane ωμ(t) — both computed from the raw 8×4 matrix. It cannot be tuned without changing the underlying scores. If you doubt the score of any (society, year, node, measure), the envelope CSVs give you the inter-rater SD; values within ±1 SD do not move κ across tiers in any society we have inspected.
How do you know the AI scorers aren't just memorising outcomes?
The protocol is blinded scoring: each year is presented to the raters without the terminal-state context. We then check inter-rater reliability — three or five independent models converge on similar scores when the structural conditions are similar, even when the public narrative is dramatic. Where they disagree (envelope SD > 1.5), that itself is a signal — the model treats high-disagreement years as low-confidence cells, not as noise to suppress.
Why eight nodes? Why not nine or fifteen?
Eight is the smallest set that admits a non-trivial three-layer decomposition with at least two members in the smallest layer. Fewer than eight collapses Mythic into a single node, which loses the Lore/Archive distinction (doctrine vs. memory). More than eight introduces nodes that load on the same factor as existing ones (e.g. "judiciary" loads heavily on Helm + Stewards + Archive and adds dimensions of variance the rater consistency cannot support). Eight is the empirical sweet spot.
What does negative Stress mean? I see S = −2 in some files.
Some published series use the signed-Stress convention (range −10 to 0) — India and Saudi Arabia in particular. The compute kernel and this guide use the absolute value. The data ingestion script flips negatives back to positive at load time so all downstream metrics are consistent.
Year transitions look smooth in the Network — is the model interpolating data?
No. The displayed values are tween-eased between two adjacent observed years for visual continuity, but every computed field (V, B, λ₂, κ, attractor) is recomputed from the interpolated matrix on each frame. The data points themselves are exactly what the ensemble produced; only the path between them is smoothed.
Which signal should I trust most when they disagree?
The composite alarm logic is designed to require corroboration: HIGH requires both κ ≥ 0.35 and helm V < 6. Single-metric alarms are downgraded to MEDIUM or LOW deliberately. If you must pick one signal, λ₂ is the most theory-grounded (it is the textbook synchronisation criterion for any networked dynamical system), and κ is the most empirically calibrated against historical collapse cases.
Can I add a new society?
Yes. Score the 8×4 matrix for each year you want, in the canonical CSV schema (Society, Year, Node, Coherence, Capacity, Stress, Abstraction). Drop it in uploads/ and ask for it to be ingested — the parser only needs the eight node names to match exactly (Lore / Archive / Helm / Stewards / Shield / Craft / Hands / Flow).
What can the framework not do?
It cannot predict which historical event will trigger a state change — only that the structural conditions for one are present or absent. It cannot resolve sub-national variation (a federation is a single matrix; regional dynamics are out-of-scope). It treats organisations and societies with the same kernel, which works at population scales but not for sub-100-person institutions where the law of large numbers fails. And it is not a forecasting tool in any conventional sense: the kernel describes the phase point of the system, not its trajectory.
How do I cite this?
The underlying framework: McKern, K. (2026). The Adaptive Cycle in Civilisational Phase Space: Mapping Holling onto the Metabolism–Myth Dyad — and the accompanying CAMS v3.2-R formulation. Datasets are CC0; cite the Wintermute repository: github.com/KaliBond/wintermute.
·Notation key
Quick lookup for symbols used throughout the apps.