Public Datasets & Transparency
Fully Open, Blinded Datasets — Scored by GPT + Grok + Gemini Ensemble. Download anything.
50
Series (43 distinct societies)
3–5
AI scorers per series (ensemble)
5 CE – 2026
Time span covered
33 datasets fully validated (Stress-column pass) · 8 further datasets undergoing quality review
Validation results
75–90%
Ensemble hindcast accuracy across tested societies
r = −0.958
Entropy–health correlation in validated datasets
50
Historical series across 43 societies, 5 CE–2026
r = 0.78
Seshat cross-validation, Latium/Rome (Apr 2026)
Methodology — how the scores are produced:
1. Blinding. Each node is scored using only information available in that historical moment — the scorer assesses function at that year without reference to what came next. Scorers are trained on historical text that may implicitly contain outcome information; this limitation is acknowledged openly in the methodology documentation.
2. Triple-AI scoring. GPT-4, Grok, and Gemini each independently assign 1–10 values for Coherence, Capacity, Stress, and Abstraction per node per time-step.
3. Ensemble average. The three scores are averaged to produce the final CAMS matrix.
4. Validation. Ensemble reliability (r > 0.7 across models) and inter-rater agreement are documented in the repository:
DATASET_VALIDATION_SUMMARY.md,
CAMS_Validation_Formulation.md.
Formal validation of the ensemble approach: The Wright Thesis — that ensemble variance reduces by exactly 1/√N and SNR improves by √5 — has been confirmed against CAMS Germany and USA data (1880–2026). Crucially, inter-rater disagreement is structured and historiographically meaningful, not random noise.
Ensemble Validation Report →
A note on ensemble scoring. Ensemble scoring was introduced to strengthen the CAMS methodology. The earliest series were scored by a single AI model, so every value carried that one scorer's reading of the historical record. An ensemble run has several scorers rate each node-year independently. It then publishes two matched files:
- the ensemble mean (ENS), which averages out any single scorer's quirks;
- the envelope (ENV), which records how far the scorers disagreed: the standard deviation of each dimension and the range of Node Value.
The envelope turns uncertainty into data, so you can see where a score is firm and where it is contested. Every ensemble on this site is paired with its envelope wherever one exists. The exceptions are listed, with reasons, in
ENVELOPE_STATUS.md. That list includes a few older single-scorer series that keep the ENS filename so the tools can still load them. All the scorers are large language models trained on overlapping text, so agreement between them shows reproducibility, not full independence.
All datasets are in the public GitHub repository under data/cleaned/. No login required.
Browse all datasets on GitHub
Full datasets index (DATASETS_INDEX.md)
New Datasets — May 2026 · cam5 Extension Series
19 nations now have matched ensemble mean + envelope pairs. Added May–June 2026: Colombia (1875–2026 annual), USA (1900–2026 5-yr), Türkiye (1875–2026 5-yr), India (1875–2026 5-yr), Japan (1875–2026 5-yr), and Australia envelope. Earlier additions: Argentina, Canada, Chile, France, UK, Germany, Norway, Poland, Russia, Sweden, Iran, China, Thailand. New series in data/nations/; earlier series in cleaned_datasets/.
Americas
Colombia — Ensemble Mean
1875–2026 · 5-Agent Ensemble · with Node Value & Bond Strength
Colombia — Envelope
1875–2026 · inter-rater SD per node per year · with V_range
United States — Ensemble Mean
1900–2026 · 5-Agent Ensemble · 5-year snapshots · with Node Value & Bond Strength
United States — Envelope
1900–2026 · inter-rater SD per node per year · with V_range
Argentina — Ensemble Mean
1950–2026 · 5-Agent Ensemble · with Node Value & Bond Strength
Argentina — Envelope
1950–2026 · inter-rater SD per node per year
Argentina
1950–2026 · 5-Agent Ensemble · raw scores (C, K, S, A) · same values as the Ensemble Mean above; shares its Envelope
Canada — Ensemble Mean
1850–2026 · 5-Agent Ensemble · with Node Value & Bond Strength
Canada — Envelope
1850–2026 · inter-rater SD per node per year · with V_range
Chile — Ensemble Mean
1875–2026 · 5-Agent Ensemble · with Node Value & Bond Strength
Chile — Envelope
1875–2026 · inter-rater SD per node per year · with V_range
Europe
France — Ensemble Mean
1850–2026 · 5-Agent Ensemble · with Node Value & Bond Strength
France — Envelope
1850–2026 · inter-rater SD per node per year · with V_range
United Kingdom — Ensemble Mean
1880–2026 · CAMNations5 · with Node Value & Bond Strength
United Kingdom — Envelope
1880–2026 · inter-rater SD per node per year · with V_range
Germany — Ensemble Mean
1880–2026 · 5-Agent Ensemble · with Node Value & Bond Strength
Germany — Envelope
1880–2026 · inter-rater SD per node per year · with V_range
Norway — Ensemble Mean
1880–2026 · 5-Agent Ensemble · with Node Value & Bond Strength
Norway — Envelope
1880–2026 · inter-rater SD per node per year · with V_range
Poland — Ensemble Mean
1875–2026 · 5-Agent Ensemble · with Node Value & Bond Strength
Poland — Envelope
1875–2026 · inter-rater SD per node per year · with V_range
Russia — Ensemble Mean
1800–2026 · 5-Agent Ensemble · with Node Value & Bond Strength
Russia — Envelope
1800–2026 · inter-rater SD per node per year · with V_range
Sweden — Ensemble Mean
1850–2026 · 5-Agent Ensemble · with Node Value & Bond Strength
Sweden — Envelope
1850–2026 · inter-rater SD per node per year · with V_range
Türkiye — Ensemble Mean
1875–2026 · 5-Agent Ensemble · 5-year snapshots · with Node Value & Bond Strength
Türkiye — Envelope
1875–2026 · inter-rater SD per node per year · with V_range
Middle East
Iran — Ensemble Mean
1875–2026 · 5-Agent Ensemble · with Node Value & Bond Strength
Iran — Envelope
1875–2026 · inter-rater SD per node per year · with V_range
Asia-Pacific
Australia — Ensemble Mean
1875–2026 · 5-Agent Ensemble · with Node Value & Bond Strength
Australia — Envelope
1875–2026 · inter-rater SD per node per year · with V_range
India — Ensemble Mean
1875–2026 · 5-Agent Ensemble · 5-year snapshots · with Node Value & Bond Strength
India — Envelope
1875–2026 · inter-rater SD per node per year · with V_range
Japan — Ensemble Mean
1875–2026 · 5-Agent Ensemble · 5-year snapshots · with Node Value & Bond Strength
Japan — Envelope
1875–2026 · inter-rater SD per node per year · with V_range
Japan — Ensemble Mean (second set, 1850–2026)
1850–2026 annual · evidence-gated 5-scorer ensemble (m5n, Sept 2026) · with Node Value & Bond Strength
Japan — Envelope (second set, 1850–2026)
1850–2026 annual · inter-rater SD per node per year · with V_range
Italy — Ensemble Mean (second set, 1850–2026)
1850–2026 annual · evidence-gated 5-scorer ensemble (m5n, Sept 2026) · with Node Value & Bond Strength
Italy — Envelope (second set, 1850–2026)
1850–2026 annual · inter-rater SD per node per year · with V_range · n_eff, NA_rate, bond_n
Netherlands — Ensemble Mean (second set, 1850–2026)
1850–2026 annual · evidence-gated 5-scorer ensemble (m5n, Sept 2026) · with Node Value & Bond Strength · one year recorded as NA (evidence too thin)
Netherlands — Envelope (second set, 1850–2026)
1850–2026 annual · inter-rater SD per node per year · with V_range · n_eff, NA_rate, bond_n
China — Ensemble Mean
1800–2025 · CAMNations5 · with Node Value & Bond Strength
China — Envelope
1800–2025 · inter-rater SD per node per year · with V_range
China — Extended Ensemble
1850–2026 · 5-Agent Ensemble · extended series
China — Envelope (1850–2026)
1850–2026 · inter-rater SD · extended series · with V_range
Thailand — Ensemble Mean
1850–2025 · 5-Agent Ensemble · with Node Value & Bond Strength
Thailand — Envelope
1850–2025 · inter-rater SD per node per year · with V_range
Ancient & Pre-Modern
Latium Vetus — Ensemble Mean · envelope missing (re-run needed)
460–2010 · Multi-Agent Ensemble · Roman institutional history
New Datasets — May 2026 · 22 Societies
22 new single-scorer and ensemble datasets added May 2026, now powering the CAMS Historical Telescope (32 societies) and fully integrated into the CAMS Explorer, CAMS Interpreter, and Zeitgeist Detector. Canonical 8-node format throughout.
Western Europe
France
1785–2024 · Single-scorer · C, K, S, A, V, BS
Netherlands
1750–2024 · Single-scorer · C, K, S, A, V, BS
Denmark
1752–2025 · Single-scorer · C, K, S, A, V, BS
Norway
1881–2025 · GEM scorer · C, K, S, A, V, BS
East Asia & Pacific
Japan
1850–2025 · Single-scorer · C, K, S, A, V, BS
Hong Kong
1900–2015 · Single-scorer · C, K, S, A, V, BS
Singapore
1935–2025 · Single-scorer · C, K, S, A, V, BS
Thailand
1850–2025 · Single-scorer · C, K, S, A, V, BS
Indonesia
1941–2025 · Single-scorer · C, K, S, A, V, BS
South Asia
India
1950–2024 · Single-scorer · C, K, S, A, V, BS
Pakistan
1947–2025 · Single-scorer · C, K, S, A, V, BS
Middle East & North Africa
Saudi Arabia
1918–2025 · Single-scorer · C, K, S, A, V, BS
UAE
1970–2026 · GEM scorer · C, K, S, A, V, BS
Iran — Ensemble Mean
5-Agent Ensemble Mean · C, K, S, A, V, BS
Iran — Envelope
Inter-rater SD per node per year
Iran — Single-scorer (no BS)
1900–2025 · C, K, S, A, V · bond strength unavailable
Iraq
1900–2025 · Single-scorer · C, K, S, A, V, BS
Israel
1946–2025 · Single-scorer · C, K, S, A, V, BS
Lebanon
1943–2025 · Single-scorer · C, K, S, A, V, BS
Syria
1893–2024 · Single-scorer · C, K, S, A, V, BS
Eurasia
Ukraine
1980–2025 · Single-scorer · C, K, S, A, V, BS
Americas & Africa
Venezuela
1970–2025 · GEM scorer · C, K, S, A, V, BS
Brazil
1880–2025 · Grok scorer · C, K, S, A, V, BS
South Africa
1880–2025 · GEM scorer · C, K, S, A, V, BS
JUNO v3 — Current Corpus (October 2026)
Which dataset to use. JUNO v3 (19 October 2026) is the current corpus and the basis for every signature result in the October 2026 thesis. The 53-society
JUNO_all_countries.csv further down is the prior corpus behind the JUNO_36 explorer; do not expect it to reproduce the signature results. Dimension definitions:
Canonical Definitions.
JUNO_Unified_Dataset_v3 — current corpus
42 societies · 41,914 node-year rows · C, K, S, A and V per node per year · basis for the frozen signature evidence in the October 2026 thesis
JUNO v3 changelog & rescore list
What was added, replaced, patched and removed from the prior corpus · series flagged for blind rescore (Denmark, Israel pre-2015, India, Turkey)
Signature Pack & thesis
The six frozen signature rules and the October 2026 thesis that reports them against JUNO v3
Quarterly runs, 2015–2026Q3
Columns: Society, Quarter, Node, then C, K, S, A, Node Value and Bond Strength. Block 1 is the ensemble mean; block 2, the envelope, gives the per-dimension SD and the V range across scorers.
United States
2015Q1–2026Q3 · 47 quarters · 8 nodes · five-scorer ensemble mean + envelope
Iran, Israel, Saudi Arabia
2015Q1–2026Q3 · 3 societies · 47 quarters each · 8 nodes · five-scorer ensemble mean + envelope
France new
2015Q1–2026Q3 · 47 quarters · 8 nodes · five-scorer ensemble mean + envelope
Germany new
2015Q1–2026Q3 · 47 quarters · 8 nodes · five-scorer ensemble mean + envelope
United Kingdom new
2015Q1–2026Q3 · 47 quarters · 8 nodes · five-scorer ensemble mean + envelope
Sri Lanka new
2015Q1–2026Q3 · 47 quarters · m5 run · 8 nodes · five-scorer ensemble mean + envelope
Targeted runs
Germany 1932–35, annual new
Weimar collapse to consolidation · 4 years · 8 nodes · five scorers
Germany 1932–35, quarterly new
1932Q1–1935Q4 · 16 quarters · 8 nodes · five scorers
USA transition quarters new
9 quarters around changes of administration, 2016Q4–2025Q2 · 8 nodes · five scorers
USA patch quarters new
Rescored quarters 2021Q2 and 2025Q1 · 8 nodes · five scorers
Telecoms, corporate nodes new
Huawei, Ericsson, Nokia, ZTE · 2010–2025 · corporate node set (Board, Risk, …) · CAMCORP5 ensemble · 512 rows
Sri Lanka raw passes new
The five individual scorer files behind the Sri Lanka quarterly ensemble
Blind rescores, v1.2-OPT (23 October 2026)
Rescores of series on the blind rescore list, scored with the CAMS raw scorer v1.2-OPT prompt. Five passes per series, run in one session by Claude Sonnet 4.6, so the passes are not independent scorers. Block 1 is the mean of the five passes, with Node Value and Bond Strength; block 2, the envelope, gives the per-dimension SD, V range, n_eff and NA rate. Not yet merged into JUNO v3.
Denmark, 1900–2026 new
Full-series rescore · 127 years · replaces the series flagged for Stress = 0 · 8 nodes · five passes
Israel, 1950–2025 new
76 years · covers the flat pre-2015 profiles flagged for rescore · 8 nodes · five passes
Ukraine, 1950–2025 new
76 years · 8 nodes · five passes
United States, 1900–2025 new
126 years · scale check against the two disagreeing USA series · 8 nodes · five passes
United States, 1900–2025 — evidence-grounded passes new
126 years · each pass takes a different lens: consensus, institutional stress, delivery record, elite coherence, Hands/Flow from below · 8 nodes · five passes
United States, 1900–2025 — substantive-variation passes new
126 years · passes agree on every dimension in 29% of rows, so this envelope runs narrower than the others · 8 nodes · five passes
JUNO 1.0 — Bond Strength Formalism
What JUNO-1.0 is: A scale-invariant bond formalism that replaces the legacy CAMS v3.2 bond formula
with a normalised alternative. Where CAMS v3.2 bonds can range up to ~64, JUNO-1.0 bonds are constrained to [0, 1],
making cross-society and cross-era comparisons directly meaningful.
Formula: Bij = √(qi · qj) · 2−(Si+Sj)/10
where qi = (0.6·C + 0.4·A) / 10
Λ (coordination index): B̄(t) / B̄0 — the system bond density at year t divided by the
society’s 90th-percentile historical SBD. Scale-invariant; comparable across all 53 societies.
Concordance with CAMS v3.2: Spearman ρ = 0.95 (SBD) and ρ = 0.95 (per-bond) across the original 36 societies —
the two formalisms agree on structural ordering while JUNO-1.0 provides a bounded, interpretable scale.
48
Societies covered
by JUNO-1.0
[0, 1]
Bond strength range
(normalised)
ρ = 0.95
Concordance with
CAMS v3.2 (orig. 36)
445 KB
Pre-computed site JSON
(all 53 societies)
Pre-computed outputs
juno_full.json — All 53 societies
JUNO-1.0 · Node Value, Bond Strength, B₀, Λ trends · Ready for site toggle · 445 KB
JUNO_all_countries.csv — Raw scores
53 societies · 47,960 rows · 20 CE–2026 · C, K, S, A per node per year · JUNO input format
juno_backcalc.py — Pipeline script
Computes JUNO-1.0 + CAMS-v3.2 node metrics, bond matrix, concordance · CLI tool · Python 3.10+
The JUNO formalism toggle is live in
CAMS Historical Telescope.
GitHub Action update_juno_data.yml auto-regenerates juno_full.json
whenever ENS source files or pipeline scripts change.
See generate_site_json.py
for the full schema specification.
JUNO v1.2-Final — Separate Run, 2026-09 Staging
Staging data — not merged into the canonical 48-society panel.
Three raw datasets run through the v1.2-Final bond-alignment kernel (the same
qi = (0.6C+0.4A)/10, Bij = √(qiqj)·2−(Si+Sj)/10
operators as JUNO_Unified_Dataset.csv) and independently re-verified with
juno/verify_bond_alignment.py: Burma (1811–2026, a society not yet in the
48-society panel) and rescores of Italy and Netherlands (1850–2026) whose
raw C/K/S/A values differ materially from what's currently published for those two societies
(1900–2024). Whether the rescores should supersede the published figures, and whether Burma
should be merged in as a 49th society, is an open methodology decision — not made here.
Every row is tagged Run = separate_run_2026-09_staging.
516
Complete society-years
(Burma 163 · Italy 177 · NL 176)
<5e-7
Max deviation vs. independent
recomputation (all asserts pass)
Files
JUNO_SeparateRun_2026-09_staging.csv
516 society-years · 4,128 rows · Node_Value_Calc, Bond_Strength_Calc, SBD_Calc, Lambda2_Calc, V_Mean_Calc, V_Min_Calc
compute_separate_run_2026-09.py
Recomputation script · documents which columns are computed vs. left not_computed and why
SeparateRun_2026-09_README.md
Provenance, what's computed vs. not, and the open merge decision
Raw source files — Italy & Netherlands rescore (1850–2026)
Unmerged with the existing published Italy/Netherlands rows in JUNO_Unified_Dataset.csv (1900–2024) — both versions remain available.
Italy — Ensemble Mean (rescore)
1850–2026 · raw Coherence/Capacity/Stress/Abstraction per node · legacy Node Value & Bond Strength columns (superseded by Bond_Strength_Calc in the staging CSV above)
Italy — Envelope (rescore)
1850–2026 · inter-rater SD per node per year · V_range/V_min/V_max/n_eff/NA_rate
Netherlands — Ensemble Mean (rescore)
1850–2026 · raw Coherence/Capacity/Stress/Abstraction per node · legacy Node Value & Bond Strength columns (superseded by Bond_Strength_Calc in the staging CSV above)
Netherlands — Envelope (rescore)
1850–2026 · inter-rater SD per node per year · V_range/V_min/V_max/n_eff/NA_rate
JUNO Blind Test Dataset — June 2026
A held-out inference challenge built on the full JUNO-1.0 corpus. Society identities are replaced with anonymous labels
(Society_A – Society_H); time is expressed as an offset (T+0 – T+226) relative to each society’s
internal baseline. All four CAMS primary metrics (C, K, S, A) are present alongside the pre-computed JUNO-1.0 derived
quantities — so the dataset can be used either as a scoring target or as a diagnostic benchmark without re-running
the pipeline. Researchers can attempt to identify the societies from structural signatures alone before consulting the
answer key (to be published separately).
8
Anonymised societies (A–H)
227
Time steps per society (T+0 – T+226)
JUNO 1.0
Bond formalism · pre-computed
Schema
| Column | Type | Description |
|---|
Society_ID | string | Anonymised label — Society_A through Society_H |
T_Offset | string | Time step relative to internal baseline (T+0, T+1, …, T+226) |
Node | string | One of the 8 CAMS nodes (Archive, Craft, Flow, Hands, Helm, Lore, Shield, Stewards) |
Coherence | float 1–10 | Internal node alignment |
Capacity | float 1–10 | Demonstrated ability to perform the function |
Stress | float 1–10 | Rate of entropy production within the function |
Abstraction | float 1–10 | Symbolic and conceptual reach |
Node_Value_Calc | float | JUNO-1.0 node value σ = A·C·(K−S) |
Bond_Strength_Calc | float [0,1] | JUNO-1.0 normalised bond strength Bij |
Lambda2_Calc | float | Λ coordination index — SBD / SBD90th |
V_Mean_Calc | float | Mean node value across all nodes at this time step |
V_Min_Calc | float | Minimum node value at this time step |
Series A
JUNO_Blind_Test.csv
8 anonymised societies (A–H) · T+0–T+226 · 9,608 rows · JUNO-1.0 derived metrics included · 621 KB · June 2026
Series B
JUNO_Blind_Test_B.csv
7 anonymised societies (A–G) · T+0–T+226 · 8,992 rows · JUNO-1.0 derived metrics included · 582 KB · June 2026
JUNO_Inflection_Points_B.csv
197 rows · detected V_Mean trajectory inflection points per society · columns: Society_ID, T_Offset, V_Mean_Calc, V_Min_Calc, V_Mean_Delta · 8 KB · June 2026
JUNO_Answer_Key_B.csv
1,124 rows · reveals real society identities, calendar years, and CAMS v1 regime classifications · columns: Society, Year, T_Offset, Society_ID, V_Mean_Calc, V_Min_Calc, CAMS_v1_Regime, Regime_Label_Calc · 80 KB · June 2026
All Datasets — 76 files · 62,871 records
| Society ↕ |
Years ↕ |
Records ↕ |
Family |
Region |
Download |
No datasets match this filter.
CSV Schema
Every dataset follows the same column structure — directly loadable into the CAMS Dashboard, Python scripts, or any data tool.
| Column |
Type |
Description |
Society | string | Society or organisation name |
Year | integer | Year of observation |
Node | string | One of the 8 institutional nodes (Lore, Archive, Helm, Stewards, Shield, Craft, Hands, Flow) |
Coherence | float 1–10 | Internal alignment and coordination clarity within the function |
Capacity | float 1–10 | Demonstrated ability to perform the function |
Stress | float 1–10 | Rate of entropy production within the function |
Abstraction | float 1–10 | Symbolic activity operationally integrated with the function (see Canonical Definitions) |
Node Value | float | Derived: C + K + (A/2) − S |
Bond Strength | float 0–1 | Derived: coupling with adjacent node |
