Neural Nations
Experimental Seshat Equinox 2022 CAMS v2.3 Cross-Dataset

Seshat × CAMS — Cross-Dataset Validation

Comparing Seshat's Social Complexity Principal Component (SPC) against CAMS cross-layer coherence Λ(t) for overlapping societies. This page tests whether two independently constructed civilisational complexity measures agree on the shape of historical trajectories.

1. Dataset Overview
Seshat Global History Databank
35
NGAs
1,494
NGA-century rows
5,500
years covered

Equinox 2022 GitHub release. SPC = 1st principal component of 8 coded complexity variables (Hierarchy, Government, Infrastructure, Information, Money, Military, etc.). MilTech = sum of 6 military technology categories. Century-resolution, −3600 to 1900 CE.

Source: Turchin et al., Seshat Global History Databank. seshatdatabank.info

CAMS v2.3 Corpus
38
Societies
39,351
Node-year records
45
Series

Annual or 5-year resolution. Λ(t) = mean cross-node bond strength = mean over all 28 node pairs of Bij(t) = √(max(Vi+8,0)·max(Vj+8,0)) / 32. Covers CE 5 – 2026, with most modern datasets starting 1750+.

Dataset gap: Seshat ends at 1900 CE (and most NGAs at 1700–1800). CAMS modern datasets begin at 1750+. The only society with direct temporal overlap is Rome / Latium (CE 0–450). The Latium/Rome comparison below is the primary cross-validation case; the broader NGA explorer shows where future CAMS reconstruction could extend the analysis.
2. Latium × Rome: Cross-Validation (CE 0–450)

Seshat codes Latium as its NGA for the Roman civilizational phase. CAMS has an independent 5-year-resolution dataset for Rome (CE 10–450) scored by Claude Sonnet 4.5 using the CAMS v2.3 rubric. Both series are century-averaged for alignment. This comparison is fully independent: Seshat uses coded historical variables; CAMS uses an LLM-based complexity rubric — no shared inputs.

Dual-Axis: CAMS Λ(t) vs Seshat SPC — Latium/Rome, CE 0–400
Century-Level Data Table
Century CECAMS Λ(t)Seshat SPCSeshat MilTech
Correlation (N = 5)
Finding
  • Both series peak at CE 100 — the height of the Flavian/Trajanic Roman Empire. Seshat SPC = 7.79 (maximum); CAMS Λ = 2.88 (maximum).
  • Both decline monotonically CE 100 → 400. CAMS Λ drops 55% (2.88 → 1.31); Seshat SPC drops only 1.5% (7.79 → 7.68) in the same window — CAMS captures internal coherence loss faster than the structural complexity index.
  • Pearson r = 0.78, Spearman ρ = 0.70 (N=5; not statistically significant — see caveat).
  • The gradient divergence is a genuine scientific signal: structural complexity (institutions, military, territory — Seshat SPC) persists longer than cross-layer coordination (CAMS Λ). This is consistent with the "hollow state" phenomenon in late Roman historiography.
Statistical caveat: N = 5 century-averaged observations. Pearson and Spearman coefficients are directionally informative but have no statistical power at this sample size. Granger causality requires N ≥ 15 after lag removal. A proper test would require CAMS reconstruction of Rome at century resolution back to BCE or CAMS forward-extension of Seshat NGAs into the 19th–20th century.
3. Seshat NGA Explorer — All 35 Natural Geographic Areas
SPC Trajectory
NGA Metadata
All NGAs — SPC at Maximum (sparkline context)
4. Proposed Seshat NGA → CAMS Society Mapping

The table below lists proposed correspondences between Seshat NGAs and CAMS societies, ordered by temporal overlap quality. "Overlap" = the intersection of Seshat and CAMS time windows.

Seshat NGABest CAMS MatchSeshat RangeCAMS RangeOverlap (N centuries)Granger-ready?
LatiumRome (recalculated)−3600 → 1800CE 10–450CE 0–400 (N=5)No — N<15
Paris BasinFrance (1785–2024)−3200 → 17001785–2024None (85y gap)No — no overlap
KansaiJapan (1850–2025)−600 → 18001850–2025None (50y gap)No — no overlap
Middle Yellow River ValleyChina (1900–2026)−2000 → 19001900–2026CE 1900 (N≈1)No — N<15
Susiana / S. MesopotamiaIraq / Iran−4000 → 19001900–2024CE 1900 (N≈1)No — N<15
Upper EgyptNo CAMS analog yet−3600 → 1700——No match
Deccan / Middle GangaIndia (if extended)−1500 → 1800——No match
Path to Granger: Two routes exist. (1) Extend CAMS backward — reconstruct Latium/Rome at century resolution from BCE to 476 CE using archival sources and the CAMS v2.3 rubric. This would yield N ≈ 40 usable Seshat-aligned centuries. (2) Extend Seshat forward — adapt the Seshat coding rubric to 20th-century polities in the CAMS corpus. Either approach would enable the full bidirectional Granger test (Λ → SPC and SPC → Λ) using granger_stationary_safe() from the validated tool suite.
5. Dataset Gap Visualisation
Temporal Coverage: Seshat NGAs vs CAMS Societies

Each bar shows the time span covered. Seshat NGAs are brown; CAMS societies are blue. The only vertical overlap is in the CE 0–450 window (Latium/Rome). This chart is the core motivation for the proposed backward-reconstruction programme.

6. Methodology
SPC — Seshat Social Complexity Index

SPC is the 1st principal component (PC1) of 8 coded complexity variables from the ImpSCDat sheet (imputed). The 8 components are:

  • PolPop — polity population
  • PolTerr — territory
  • CapPop — capital population
  • levels — administrative hierarchy levels
  • government — government specialisation
  • infrastr — infrastructure
  • writing — writing/information
  • money — monetary system

Higher SPC = more structurally complex polity. Range ≈ 2.3–8.3 in this corpus.

Λ(t) — CAMS Cross-Layer Coherence
Vi(t) = Ci + Ki + Ai/2 − Si Bij(t) = √(max(Vi+8,0)·max(Vj+8,0)) / 32 Λ(t) = mean over all 28 node pairs of Bij(t)

C = Coherence, K = Capacity, A = Abstraction, S = Stress. 8 nodes × 4 metrics. Bond strength Bij captures the geometric mean of two nodes' adjusted values. Λ(t) = mean coupling across all node pairs.

Range in Rome corpus: 0.73 (CE 500, terminal decline) to 2.97 (CE 100, peak).

Why the two metrics should correlate

Both SPC and Λ(t) are designed to capture civilisational complexity, but from different theoretical angles. SPC captures structural endowment — what institutional and material resources a polity possesses. Λ(t) captures coordination quality — how well those resources are being integrated across functional layers.

The finding that Λ(t) declines faster than SPC during Rome's CE 100–400 period is theoretically meaningful: structural endowment persists (armies, roads, laws remain nominally intact) while cross-layer coordination degrades (fiscal stress, elite conflict, provincial fragmentation). This is precisely the CAMS prediction: coordination failure precedes structural collapse.

The Granger question — whether Λ(t) leads SPC decline, or vice versa — remains open pending sufficient overlap data.