The structural pass
Version 0.1 is a structural hypothesis. Its nodes are the 73 capability groups of the National Capability Framework, in nine domains across hard, soft and economic capability, taken unchanged from the framework’s published node sheet. Its edges were written by hand, one at a time, from two sources: the framework’s own definitions of each group and the seams between them, and the published literature on how capabilities depend on each other. That literature includes economic complexity and capability adjacency (Hidalgo and Hausmann), national innovation systems (Lundvall, Nelson, Freeman), infrastructure and growth (the World Bank, the OECD, Calderón and Servén), human capital and growth (Hanushek and Woessmann), institutions and governance (Acemoglu and Robinson, the Worldwide Governance Indicators literature) and critical technology and security dependencies (ASPI, CSET). Each edge cites the literature area it draws on, not a specific paper: the structural pass asserts a direction and a rough size, and does not claim that any one study estimated it.
Edges between domains were preferred. An edge inside a domain was kept only where the dependency is strong and well known, such as foundational education into advanced education, or energy infrastructure into advanced manufacturing.
What an edge says
An edge from A to B says that a change in A tends to produce a change in B. It carries:
- Type. Enabling: B rises with A. Substituting: A stands in for B, so B falls as A rises (allied guarantees reduce the pressure to raise national forces). Competing: A and B draw on the same resources (a resource boom crowds out diversification).
- Weight, from 0 to 1: the share of B’s change that runs through this edge. A weight of 0.3 says that about three tenths of the movement in B is explained by movement in A.
- Elasticity, as a bracket: the per cent change in B for a 1 per cent change in A, from a lower to an upper bound. The graph publishes the bracket, never a single point.
- Lag: the years before the effect begins, the years until it is complete, and the shape in between. A linear ramp builds evenly; a geometric lag front-loads the effect, closing a constant share of the remaining gap each year.
- Confidence (low or medium in this version), provenance (structural for every edge in this version), a one- or two-sentence rationale and the references.
Lags follow the channel the source works through: schooling and early childhood 5 to 15 years; tertiary education, skills and research 3 to 10; institutions 2 to 10; health 2 to 8; infrastructure 1 to 5, or 2 to 7 for slow-built assets; technology 1 to 4; forces 1 to 5; finance and budgets 0 to 2, or 1 to 3 into capital stock; trade and investment 0 to 3. Where a fast channel feeds something that has to be built, raised or taught, the edge takes the slower lag of what it builds: a budget moves within a year, but the grid, the regiment or the school it pays for does not.
Reading weight and elasticity together
Weight and elasticity answer different questions. Weight is how much of the target’s movement runs through the edge; elasticity is how strongly that channel transmits a change. In propagation an edge passes on weight × elasticity times each change in its source, spread over the lag. A sustained 10 per cent rise in a source with an edge of weight 0.3 and elasticity 0.4 to 0.6 raises the target by 1.2 to 1.8 per cent once the lag has run, before any knock-on effects through other edges.
Normalisation
The weights into each group are capped so that together they explain at most 0.9 of its change. At least a tenth is left to the group’s own momentum: the part of its movement the graph does not explain. Where a group’s incoming weights in the seed add up to more than 0.9, all of them are scaled down in proportion at build time; the seed’s own values are kept beside the normalised ones in the published table.
Checks every version passes
The build fails, and reports every failure at once, if any edge names a group that does not exist, links a group to itself or repeats another edge; if the weights into any group sum to more than 1; if an elasticity bracket or a lag runs backwards; if an edge has no rationale or no references; or if the spectral radius of the weighted adjacency matrix is 1 or more. Below 1, a shock dies away as it passes round the graph’s cycles and propagation converges; at 1 or above it would grow without limit. The build also reports the heaviest cycles, and warns, without failing, about any group with no incoming or no outgoing edge.
What comes next
The structural pass is the first of three. An empirical pass will estimate lagged relationships between group scores over time from the GINC Research, Patents, Trade and Infrastructure indices, and confirm, rescale or contradict structural edges. An expert pass will put the edges to the GINC expert network through a cross-impact balance elicitation, with each expert rating the direction and strength of influence between pairs of groups. Edges supported by both will be able to reach high confidence, which the structural pass cannot award itself. Nation-specific values, where the dependency is known to differ for a particular nation, will follow as overrides of the global graph.