This playground is a sandbox for reasoning about the Great Divergence, the opening of a large per-capita income gap between an industrialising "North" and the rest of the world over the long nineteenth and twentieth centuries. It does not reconstruct GDP. Instead it asks a methodological question: once you decide which factors mattered, how do you combine them into a single regional "development score", and how do you fairly attribute the resulting gap among those factors? The model exposes three aggregation rules and a cooperative-game attribution method, and it is deliberate about the fact that the attribution depends on the aggregation rule you pick. There is no model-free credit split.
Long-run reconstructions of per-capita output, most prominently the Maddison Project, support a now-standard stylised picture. Before roughly 1750 the richest and poorest large regions of the world differed in income by a modest factor, perhaps two to three. By the mid-twentieth century that factor had grown to something like ten or more. The gap is therefore mostly a modern construction, coincident with industrialisation, fossil-energy throughput, imperial integration of global markets, and the institutional and human-capital transformations that accompanied them. This is the sense in which Kenneth Pomeranz titled his 2000 book The Great Divergence: the interesting fact is not that some regions were always ahead, but that a wide gap opened in a specific historical window.
The literature offers competing emphases rather than a single cause:
These accounts are not mutually exclusive. They disagree about weights and about which factors are proximate versus fundamental. The playground takes no side. It lets the user encode a weighting and then shows the consequences.
The model carries nine accelerants: energy throughput, institutions and property rights, state capacity, human capital, knowledge and innovation, finance and capital markets, trade and value-chain power, coercion and empire, and geography and disease. Each accelerant is reduced to a single value in the unit interval per region per time bin. This is a severe reduction, and the playground says so. The whole point of the exercise is to make the reduction explicit and then reason carefully inside it.
The default timeline has ten bins, from early agrarian states (before year 1) through industrial takeoff (1750 to 1850) to the post-2008 multipolar period. The default North and South values are illustrative, not measured. They are set to tell the stylised story: near-parity in antiquity, a widening gap from industrial takeoff onward, and partial catch-up after 1950. Users are expected to override them with their own estimates.
A region's composite score is computed from its accelerant values and a vector of weights. The weights are renormalised to sum to one, so the composite is a proper weighted mean. Three aggregators are available:
Additive, the weighted arithmetic mean, treats accelerants as perfect substitutes. A high energy score can fully compensate for weak institutions.
f = sum_i w_i x_i
Multiplicative (Cobb-Douglas), the weighted geometric mean, gives unit elasticity of substitution. A factor near zero drags the whole product down: you cannot industrialise on energy alone if institutions are absent.
f = product_i x_i^(w_i)
CES, the constant-elasticity-of-substitution form, has a tunable parameter rho that nests the other two. At rho = 1 it is exactly the additive form; as rho approaches 0 it approaches the geometric mean; negative rho makes accelerants complements, so the weakest factor dominates (a weakest-link regime).
f = (sum_i w_i x_i^rho)^(1/rho)
The choice of aggregator is a hypothesis about how development factors combine, and it is the single most consequential modelling decision in the playground. The calibration panel verifies that these limit identities hold exactly: the CES at rho = 1 matches the additive form, and the multiplicative branch matches an independently computed geometric mean.
The North-South gap is reported in one of two modes. Difference mode returns North score minus South score; ratio mode returns North score over South score. In the default timeline the difference is essentially zero in antiquity (the two regions share the same illustrative values) and grows positive through the industrial era. This is the toy's encoding of the Great Divergence stylised fact.
Given a fixed aggregator and a fixed bin, the model attributes the gap among the nine accelerants using the Shapley value from cooperative game theory. The Shapley value is the unique fair allocation satisfying efficiency, symmetry, the null-player axiom, and additivity: each accelerant gets its average marginal contribution to the gap over all possible orderings in which factors are switched on.
phi_i = average over orderings of [ v(S before i, with i) minus v(S before i) ]
Here the characteristic function v is the gap under the chosen aggregator when only a subset of accelerants is "active". Because enumerating all nine-factorial orderings is expensive, the model estimates the Shapley values by Monte Carlo sampling of permutations (200 by default), with a deterministic seed per bin so the result is reproducible.
The crucial honest point: the attribution is conditional on the aggregator. Under the additive form, factor interactions vanish and the Shapley split reduces to the weighted contribution of each factor's North-South difference. Under CES or Cobb-Douglas, interactions are real, and the same data can reorder the credit. There is no model-free decomposition of cause. The Shapley value is a fair split of a chosen game, not a discovery about history.
The model reproduces, by construction, the stylised divergence trajectory baked into its default values. It does not derive that trajectory from data, and it should not be read as evidence for any particular causal story. Its genuine content is methodological: