Batch mcts-m7. Model 0.2.0, graph 0.2.0, parameters 0.1.1. 10 simulations, seeds 1 to 10; probes on 3. Every figure is a distribution over simulations of a structural model with seed parameter values; it shows what the model implies under these settings, not what will happen. Other nations are frozen at their 2026 ratings. No reconstructed historical data is used in this experiment.
Headline
South Korea played 30 rounds from 1996 under the tree search (mcts) method, with scripted events, 10 times. Its capability index moved from 13.13 to a median of 15.13 (+2.00), with 90 per cent of simulations between 15.11 and 15.18 [experiment_objectives].
Starting position
In 1996 the nation ranks 21 on the index against the frozen field, holds 0 of nine domains at Tier 1 and has a tier rank of 22. Its base income is 649 credits a year, calibrated so that its 2026 position is affordable with headroom for investment.
What was decided
A simulation spent a median of 3,100 credits on actions. The five groups that took most were Fiscal Capacity & Public Finance (partner, 170 credits, 5.5 per cent); Space Technologies (partner, 166 credits, 5.4 per cent); Investment Attraction & Projection (partner, 118 credits, 3.8 per cent); Artificial Intelligence (buy, 104 credits, 3.3 per cent); Economic Leverage & Dependence (partner, 98 credits, 3.1 per cent) [experiment_decisions_by_group]. By mode: partner 42 per cent, buy 30 per cent, build 28 per cent [experiment_decisions_by_group]. Concentration of spending (HHI) had a median of 0.029, and the most repeated action appeared in 30 per cent of rounds.
The leading domain by period: rounds 1 to 5, Human Capital; rounds 6 to 10, Human Capital; rounds 11 to 15, Financial Strength; rounds 16 to 20, Critical Technology; rounds 21 to 25, Critical Technology; rounds 26 to 30, Critical Technology [experiment_spend_by_period].
The method's chosen action was within 5 per cent of its best alternative in 97 per cent of actions and ahead by more than 25 per cent in 0 per cent [experiment_decision_margins].
What the world did
A median simulation saw 4 shocks and 0 upside events. The most damaging per firing were S10 (10 firings in 10 simulations, −23.4 points each); S02 (20 firings in 10 simulations, −15.7 points each); S01 (10 firings in 10 simulations, −12.6 points each) [experiment_shocks].
The groups that lost most to shocks, per simulation: Investment Attraction & Projection −2.22; Fiscal Capacity & Public Finance −2.21; Labour & Employment −2.10; Monetary & Currency Capability −2.02 [experiment_shock_damage_by_group].
The edges that carried most change through the graph, per simulation: Space Technologies to Space Operations (+1.39 points); Space Technologies to Strategic Strike & Missile Defence (+1.30 points); Fiscal Capacity & Public Finance to Social Infrastructure (+0.67 points) [experiment_propagation_by_edge].
Where it ended
At 2026 the median rank is 10 (5th to 95th percentile 10 to 10), the median count of Tier 1 domains 3 (2 to 4) and the median tier rank 5; the resilient score, the mean score sum over simulations, is 1,100 [experiment_objectives].
Tier 1 probability changed most in: financial strength (0 to 70 per cent of simulations); human capital (0 to 100 per cent of simulations); trade and investment (0 to 100 per cent of simulations) [experiment_tiers_by_round].
Why
Without shocks the index ends +1.025 higher on average, so the shocks cost about 1.025 index points [experiment_probe_attribution].
Holding instead of following this method changes the index by −3.244: that is the method's worth against doing nothing [experiment_probe_attribution].
Removing any single one of the five largest actions changes the index by between −0.025 and −0.014, for example 2020 buy Biotechnology & Health Technologies (23 credits) [experiment_probe_attribution]. The largest actions by credits come late, when income is highest, and are rarely the most valuable.
Removing the largest shock of a simulation raises the index by +0.349 on average [experiment_probe_attribution].
The replay probe reproduces the simulations exactly (index change 0.000000), so the attributions are clean [experiment_probe_attribution].
Surprises
tier_jump (warn): largest gain in one domain within five years: 7 tiers (10 simulations above 2); tiers are reported, not an objective.
The simulation furthest from the median (seed 10) ends +0.05 from it, with 4 shocks and 0 upside events [experiment_outlier_simulations].
Comparison
This is the baseline (hold) experiment, or there is none for this nation in the batch.
No backcast comparison: this experiment does not start from a historical position [experiment_against_history].
The search
Monte Carlo tree search 0.1.0 with candidate generators 0.1.0 (marginal return, graph leverage, goal gap; top 4 groups each), 500 iterations a decision, UCT c = 1.4 scaled to the reward spread, widening alpha = 0.5, greedy playouts to the horizon, objective index; playouts drew their futures from the none (certainty-equivalent playouts: the planner ignores shocks) scenario.
The committed decision most often chosen, round by round: round 1 partner Digital & AI Competence (30 per cent of simulations, visit share 31 per cent); round 2 partner Research Talent & Knowledge Creation (30 per cent of simulations, visit share 11 per cent); round 3 greedy (40 per cent of simulations, visit share 9 per cent); round 4 greedy (40 per cent of simulations, visit share 23 per cent); round 5 greedy (20 per cent of simulations, visit share 19 per cent) [mcts_decision_sequence]. The search hesitated (top two root candidates within 10 per cent) in 54 per cent of decisions, considered a mean of 32 candidates, grew trees of a mean 458 nodes to depth 4.9, and took 1.2 seconds a decision [mcts_decision_sequence].
Where the committed actions came from, by proposing generator (committed simulation-rounds): marginal return 18.0; greedy 8.8; graph leverage 4.9; graph leverage | marginal return + graph leverage 2.6; marginal return + graph leverage | graph leverage 2.1 [mcts_root_candidates].
Against the greedy experiment with the same seeds (m7-kor-greedy), the search's index at the horizon is +0.042 on average (10th to 90th percentile +0.015 to +0.069), better in 100 per cent of seeds; on the index +0.042, the score sum +2.7 and Tier 1 domains +0.40 [mcts_regret].
Across seeds the same root decision won in a mean 41 per cent of simulations (first five rounds 32 per cent) [mcts_stability].
The principal variation was followed 100 per cent at depth 1, 45 per cent at depth 2, 30 per cent at depth 3, 31 per cent at depth 4 [mcts_pv_vs_path].