C1 · cmaes-c1.c1-kor-cmaes-allocation

South Korea, evolution strategy, 20 years

5 simulations of 20 rounds from the 2026 position. South Korea decides by the evolution strategy (cma-es) method; the shock and upside library, drawn at random each round.

Model 0.2.0 · graph 0.2.0 · parameters 0.1.1 · seeds 1–5 · elasticity sampled within each bracket

Every simulation · Compare with another experiment · More experiments like this

Not a forecast

Simulations of a structural model with seed parameter values, to compare methods and test the model. Other nations stay at their 2026 ratings throughout.

Outcome across simulations

The median at 2046, with the 10th to the 90th percentile of the 5 simulations under it.

Capability index: 18.07 (+2.06); From 16.00; 17.39 to 18.25
Rank by index: 1; From 6; 1 to 2
Domains at Tier 1: 7; From 2; 6 to 7
Tier rank: 2; From 7; 2 to 2

Capability index by round

Rank by index, by round

Every objective at the position and at the horizon
Objective	2026	10th percentile	Median	90th percentile	Better
Capability indexprimary	16.00	17.39	18.07	18.25	higher
Rank by index	6	1	1	2	lower
Tier rank	7	2	2	2	lower
Domains at Tier 1	2	6	7	7	higher
Score sum	1165.8	1255.3	1309.1	1320.4	higher
Goal-weighted score	1165.8	1255.3	1309.1	1320.4	higher
Resilient score	1165.8	1292.9 (mean score sum over the simulations)			higher

The capability index is the mean of the three dimensions, each the mean of its domains, each the mean of its groups. Tiers are Pareto layers within each domain; the tier rank orders nations by domains held at Tier 1, then Tier 2, and so on. The goal-weighted score is the score sum unless the experiment sets a goal. The primary objective is the one the experiment is judged on; every other objective is reported beside it.

Credits

What each round’s credits went on in the median simulation. Across all simulations, build took 59 per cent, buy took 15 per cent, partner took 26 per cent of the 2,897 credits a simulation spent on actions.

Simulations

Three simulations told round by round, each on its own page: the one nearest the median index at the horizon, and those nearest the 10th and the 90th percentile. Every one of the 5 is listed with its outcome on the simulations page.

All 5 simulations

Events

How often each shock and upside event began across the simulations. Shocks come from the ten downside scenarios and upside events from the ten upside ones, at their yearly likelihood.

Event	Kind	Simulations	Mean round	Points
Regional war spillover S03	Shock	40 per cent	13.7	−30.0
Breadbasket failure S06	Shock	40 per cent	2.0	−13.5
AI labour shock S08	Shock	40 per cent	3.5	−21.7
Market access opening S12	Upside	40 per cent	5.0	+15.2
Water-efficient food leap S16	Upside	40 per cent	12.5	+15.1
Automated discovery step S17	Upside	40 per cent	16.5	+40.8
Hybrid coercion S04	Shock	20 per cent	7.0	−13.7
Core disaster S07	Shock	20 per cent	17.0	−16.2
Middle East settlement S14	Upside	20 per cent	9.5	+15.4
Clean power threshold S15	Upside	20 per cent	16.0	+32.9
Chronic disease advance S19	Upside	20 per cent	6.0	+15.2

Groups across simulations

Every capability group: its position, and its median and spread at the horizon across the simulations, with what the method spent on it.

Find a group

Domain

73 of 73 groups

Capability group	Domain	2026	2046	Change	Levels invested	Credits
Resource Infrastructure	Strategic Infrastructure	11.75	16.84 14.57 to 17.03	+5.09	4.21	93
Transparency, Oversight & Accountability	Governance & Integrity	15.33	20.00 19.93 to 20.00	+4.67	4.75	127
Technical & Vocational Skills	Human Capital	15.50	20.00 19.88 to 20.00	+4.50	5.52	67
Public Integrity & Anti-Corruption	Governance & Integrity	15.00	19.10 18.21 to 19.67	+4.10	3.35	87
Diplomatic Statecraft & Global Networks	Information & Influence	14.50	18.58 18.23 to 18.81	+4.08	4.04	80
Development Cooperation & Global Assistance	Information & Influence	14.50	18.56 18.35 to 18.94	+4.06	4.10	73
Lifelong Learning & Reskilling	Human Capital	16.00	19.99 19.82 to 20.00	+3.99	5.80	79
Economic Leverage & Dependence	Trade & Investment	15.33	19.16 17.37 to 19.31	+3.83	3.59	60
Alliances & Security Cooperation	Information & Influence	16.00	19.80 18.79 to 19.95	+3.80	3.76	55
Justice System & Fundamental Rights	Governance & Integrity	15.00	18.79 18.47 to 19.63	+3.79	3.44	97
Information Environment & Strategic Communication	Information & Influence	15.67	19.44 17.77 to 19.78	+3.77	3.48	83
Political Institutions & Civic Participation	Governance & Integrity	16.00	19.71 19.59 to 20.00	+3.71	3.51	93
Regulatory Governance & Strategic Controls	Governance & Integrity	16.33	19.93 19.17 to 20.00	+3.60	3.19	99
Business Dynamism & Entrepreneurship	Production & Innovation	15.75	19.23 18.25 to 19.41	+3.48	3.05	67
Sovereign Capital & External Financial Power	Financial Strength	14.20	17.45 16.73 to 18.35	+3.25	3.44	73
Population Health	Human Capital	16.00	19.22 18.66 to 19.61	+3.22	2.89	66
Financial Regulation & Integrity	Financial Strength	15.75	18.83 18.35 to 19.79	+3.08	3.35	86
Investment Attraction & Projection	Trade & Investment	16.33	19.37 18.73 to 20.00	+3.04	3.14	60
Labour & Employment	Production & Innovation	15.00	18.02 17.36 to 18.60	+3.02	2.94	60
Digital & AI Competence	Human Capital	17.00	19.85 19.06 to 20.00	+2.85	2.78	49
Economic Output & Productivity	Production & Innovation	16.50	19.26 18.57 to 19.38	+2.76	2.54	66
Advanced Education	Human Capital	17.00	19.67 19.63 to 19.92	+2.67	2.52	61
Industrial & Sectoral Base	Production & Innovation	16.50	19.16 18.78 to 19.79	+2.66	2.86	73
Fiscal Capacity & Public Finance	Financial Strength	16.50	18.97 17.45 to 19.30	+2.47	2.37	92
Research Talent & Knowledge Creation	Human Capital	16.00	18.35 17.68 to 19.18	+2.35	1.66	43
Structural Transformation & Complexity	Production & Innovation	16.67	19.00 18.68 to 19.45	+2.33	2.38	66
Health System Capability	Human Capital	16.67	18.97 17.47 to 19.41	+2.30	1.48	46
Foundational Education	Human Capital	17.25	19.49 18.96 to 19.72	+2.24	1.90	49
Financial Infrastructure	Strategic Infrastructure	16.50	18.68 17.69 to 18.82	+2.18	1.69	40
Government Effectiveness & Public Administration	Governance & Integrity	17.20	19.32 17.84 to 19.82	+2.12	0.95	31
Digital & Services Trade	Trade & Investment	16.25	18.29 17.42 to 19.47	+2.04	2.44	51
Diaspora & Global Communities	Information & Influence	15.50	17.46 16.74 to 18.17	+1.96	2.17	51
Culture, Language & Education Diplomacy	Information & Influence	16.80	18.62 18.46 to 19.17	+1.82	1.91	52
Artificial Intelligence	Critical Technology	16.00	17.79 17.56 to 18.27	+1.79	1.88	31
Space, Aerospace & Defence Infrastructure	Strategic Infrastructure	16.33	18.06 17.39 to 18.66	+1.73	1.48	47
Connectivity, Cyber & Trust Technologies	Critical Technology	16.60	18.32 18.04 to 18.66	+1.72	1.84	33
Sensing, Timing & Navigation	Critical Technology	15.80	17.35 16.82 to 17.90	+1.55	1.50	33
Space Technologies	Critical Technology	14.67	16.22 15.89 to 16.75	+1.55	1.59	44
Monetary & Currency Capability	Financial Strength	16.33	17.80 16.41 to 18.91	+1.47	1.54	45
Financial System Depth & Access	Financial Strength	17.00	18.27 18.06 to 19.38	+1.27	1.64	63
Export Capability & Sophistication	Trade & Investment	17.00	18.27 17.53 to 19.03	+1.27	1.25	24
Early Childhood Development	Human Capital	17.00	18.26 17.73 to 18.95	+1.26	0.84	19
Innovation & Commercialisation	Production & Innovation	18.00	19.25 17.74 to 19.40	+1.25	0.10	7
Trade Facilitation & Logistics	Trade & Investment	17.50	18.71 17.24 to 19.37	+1.21	1.51	40
Trade Policy & Market Access	Trade & Investment	17.00	18.18 17.49 to 19.95	+1.18	1.34	25
Social Infrastructure	Strategic Infrastructure	17.00	18.08 16.37 to 18.29	+1.08	0.00	0
Internal Security & Order	Governance & Integrity	15.67	16.72 16.50 to 17.44	+1.05	0.69	19
Advanced Manufacturing	Strategic Infrastructure	18.40	19.40 17.90 to 19.54	+1.00	0.00	0
Biotechnology & Health Technologies	Critical Technology	16.20	17.14 16.53 to 17.70	+0.94	0.51	13
Transport Infrastructure	Strategic Infrastructure	16.50	17.38 15.25 to 17.74	+0.88	0.00	0
National Resilience & Crisis Management	Governance & Integrity	16.20	17.07 15.41 to 18.70	+0.87	0.63	19
Global Value Chain Position	Trade & Investment	17.33	18.17 17.67 to 19.00	+0.84	1.17	32
Communications & Digital Infrastructure	Strategic Infrastructure	18.00	18.80 17.94 to 18.93	+0.80	0.27	6
Energy Infrastructure	Strategic Infrastructure	16.13	16.92 14.96 to 18.43	+0.79	0.68	20
Water, Sanitation & Waste Infrastructure	Strategic Infrastructure	16.80	17.47 16.60 to 17.58	+0.67	0.22	6
Advanced Materials & Manufacturing	Critical Technology	16.71	17.23 16.52 to 17.89	+0.52	0.22	7
Quantum Technologies	Critical Technology	14.25	14.77 14.10 to 15.38	+0.52	0.21	7
Financial Stability & Crisis Management	Financial Strength	16.67	17.19 16.84 to 18.49	+0.52	0.87	27
Technology Adoption & Diffusion	Production & Innovation	18.00	18.52 17.68 to 19.04	+0.52	0.00	0
Agriculture & Food Infrastructure	Strategic Infrastructure	15.75	16.20 15.38 to 16.42	+0.45	0.20	6
Energy Technologies	Critical Technology	16.60	17.04 15.94 to 17.93	+0.44	0.20	7
Semiconductors & Advanced Computing	Critical Technology	17.50	17.93 17.23 to 17.96	+0.43	0.01	0
Robotics & Autonomous Systems	Critical Technology	16.50	16.92 16.29 to 17.19	+0.42	0.00	0
Air Power	National Security	16.75	17.03 15.84 to 17.17	+0.28	0.00	0
Cyber & Electromagnetic Operations	National Security	16.00	16.18 14.95 to 16.42	+0.18	0.00	0
Force Generation & Mobilisation	National Security	16.33	16.49 15.40 to 17.58	+0.16	0.36	13
Space Operations	National Security	13.67	13.83 12.81 to 14.69	+0.16	0.18	6
Strategic Strike & Missile Defence	National Security	16.00	16.15 14.37 to 16.83	+0.15	0.20	7
Intelligence & Assessment	National Security	15.83	15.94 14.77 to 16.50	+0.11	0.00	0
Joint Command, Projection & Sustainment	National Security	15.33	15.43 13.53 to 16.21	+0.10	0.20	6
Maritime Power	National Security	16.00	16.07 14.03 to 16.30	+0.07	0.00	0
Land Power	National Security	16.50	16.53 16.04 to 17.40	+0.03	0.37	13
Nuclear Deterrence	National Security	1.67	1.66 1.45 to 2.28	−0.01	0.00	0

The optimisation

Capability index at 2046, against the greedy experiment with the same seeds and, where one exists, the tree search.

Capability index, median: 18.07; From 16.00 at the 2026 position; 17.39 to 18.25
Regret against greedy: +0.071; ahead in 60% of seeds; 10th to 90th −0.060 to +0.181
Against the tree search: +0.046; ahead in 80% of seeds
Hold-out gain: +0.1589; of the best plan over the greedy start, mean over optimisations

Objective: Capability index at the horizon (higher is better)
Encoding: Allocation shares: 4,400 numbers over the full horizon, separable (diagonal) covariance
Mode: Open-loop: optimised once from the position over 20 rounds, then played with the realised shocks
Budget: population 29; up to 45 generations; stagnation stop at 30. Ran a mean of 45 generations (5 on budget)
Shock seeds: 8 a generation, every plan of the generation on the same seeds (common random numbers), redrawn each generation; 8 fixed hold-out seeds for the curve. Plans scored under the stochastic scenario.
Start and limits: Started from the greedy plan, encoded; sigma 0.3; a single action capped at the large chunk (5 per cent of income), as greedy
Pairs: Greedy: the same seeds. Tree search: m2b-kor-mcts-index (mcts-m2b), the same nation, horizon, objective and seeds.
Versions: CMA-ES 0.1.0, the cma package 4.5.0
Runtime: 42 seconds of optimisation a simulation

Compare with another CMA-ES experiment · JSON · The method · The tree-search pair

Capability index by round

The median across the 5 simulations over the band from the 10th to the 90th percentile, with the greedy experiment's median as the dashed reference. Under the axis, the share of simulations a shock struck in each round.

shocks

Tree search, median and 10th to 90th percentileGreedy, median, same seeds

Regret against greedy

The plan's objective minus greedy's, paired by seed, by round: the median over the 10th-to-90th band. Positive is better for CMA-ES whatever the objective.

Convergence

The one optimisation, averaged over simulations: the population's best and mean fitness by generation, the hold-out fitness of the best plan so far (on fixed seeds, so it only rises), and sigma, the step size, on the right axis.

Best of the populationMean of the populationHold-out fitness of the best plan so farSigma (right axis)

Planned against realised

The credits the plan intended for each round, by domain, beside the credits played after shocks. 77% of realised actions differed from the plan.

Planned

Credits the first plan intended, by round and domain, averaged over simulations.

Realised

Credits played after shocks. Under the axis, the share of simulations a shock struck in each round.

shocks

Critical Technology
Strategic Infrastructure
National Security
Human Capital
Information & Influence
Governance & Integrity
Financial Strength
Production & Innovation
Trade & Investment

Allocation stream

Credits spent on actions by round, stacked by domain or by mode, averaged over simulations.

Tier ribbon

For each domain, the share of simulations in each tier by round.

Critical Technology
Strategic Infrastructure
National Security
Human Capital
Information & Influence
Governance & Integrity
Financial Strength
Production & Innovation
Trade & Investment

Tier 1Tier 2Tier 3Tier 4 or below2026 to 2046, left to right

Three methods compared

CMA-ES, the tree search and greedy on the same nation and objective: the regret curves overlaid, the end states by domain on one radar, and how alike their allocations are, round by round.

Regret curves, overlaid

CMA-ES minus greedy (blue) and CMA-ES minus the tree search (violet), the median paired difference in capability index by round, with the 10th-to-90th band.

End states by domain

The nine domain means at the horizon (the median over simulations) under each method, and the position.

PositionCMA-ESTree searchGreedy

Allocation similarity by round

Cosine similarity of the credits by group and mode, averaged over simulations: 1 is the same allocation, 0 none in common. Darker is more alike.

Pair	26	27	28	29	30	31	32	33	34	35	36	37	38	39	40	41	42	43	44	45	Mean
CMA-ES · greedy																					0.55
CMA-ES · tree search																					0.25
Tree search · greedy																					0.42

Attribution

From the counterfactual probes on the first 2 seeds: the position, plus the worth of the five largest actions, minus the cost of the three largest shocks, plus everything else, to the horizon. In capability index units.

20262026 16.002
D12039 partner Culture, Language & Education Diplomacy (33 credits) +0.0195
D22039 build Labour & Employment (33 credits) +0.0201
D32039 build Transparency, Oversight & Accountability (33 credits) +0.0217
D42039 build Industrial & Sectoral Base (33 credits) +0.0165
D52039 build Resource Infrastructure (33 credits) +0.0199
S12039 S03 (-40.8 points) −0.4429
S22038 S03 (-27.0 points) −0.2965
S32033 S04 (-13.7 points) −0.1995
OtherEverything else +2.2754
20462046 17.436

Fragility at the horizon

Groups of the representative simulation's end state within half a point of changing their domain's tier.

Force Generation & MobilisationNational Security · 16.5 · Tier 3+0.5 lifts a tier
Air PowerNational Security · 17.0 · Tier 3+0.5 lifts a tier

The plan, round by round

The representative simulation (seed 4, nearest the median at the horizon): the actions played each round and what the plan intended for them.

Round

Group	Mode	Played (credits)	Planned	Differs
Advanced Education	build	30.4	30.4	no
Technical & Vocational Skills	build	30.4	30.4	no
Lifelong Learning & Reskilling	build	30.4	30.4	no
Research Talent & Knowledge Creation	buy	30.4	30.4	no
Fiscal Capacity & Public Finance	partner	30.4	30.4	no

Report, with the optimisation's section

Written from the simulation record by the loop's narrator, with the planner's own section at the end. Every figure names the query it came from, in square brackets.

Settings and versions

Experiment cmaes-c1.c1-kor-cmaes-allocation. Every simulation records the model, graph and parameter versions, the scenario, the method and its seed, so it can be reproduced exactly.

Nation: South Korea (KOR)
Position: The 2026 ratings of the National Capability Index
Method: Evolution strategy (CMA-ES). An evolution strategy (CMA-ES) over a whole plan, scored by playing it through the engine under shared shock seeds: open-loop (optimised once from the position) or receding-horizon (re-optimised every round).
Scenario: The shock and upside library, drawn at random each round
Objective: Capability index, the primary objective the experiment is judged on; every other objective is reported beside it
Horizon: 20 rounds, 2026 to 2046
Simulations: 5 seeds, 1 to 5. Every method in the batch faces the same draws for the same seed.
Graph elasticities: Drawn for each simulation, uniformly within each edge’s bracket
Income: Base credits 1,014.1 a round, moving with the economic groups
Versions: Model 0.2.0, Capability Dependency Graph 0.2.0, parameters 0.1.1
Batch: CMA-ES C1, encoding comparison (cmaes-c1)
Runtime: 294.6 seconds in all; 2109553 microseconds a round

Not a forecast: Every figure is a distribution over the simulations of a structural model, reported to compare methods and test the model. None is a prediction of what a nation will do or become.
Structural parameters: Costs, decay, vulnerability and credits are structural seed values with low or medium confidence, set by domain with single-group exceptions. Five groups have no defensible value in a single-nation model and are flagged in the parameter tables.
Other nations are frozen: One nation is simulated at a time. Every other nation stays at its 2026 rating, so rank and tiers are measured against a still field; where the simulated nation rises past others their own tiers are not re-sorted.
Starting depth and mode are defaults: The position is the National Capability Index ratings. How institutionalised each capability is, and how much was built, bought or partnered, have no data behind them yet and start from structural defaults.
Credits are calibrated, not measured: Each nation’s income is set so that its position is affordable with headroom that rises with its fiscal capacity. It is not yet fiscal capacity as a share of GDP.