Batch mcts-dev. Model 0.1.0, graph 0.2.0, parameters 0.1.0. 2 simulations, seeds 1 to 2; probes on 1. 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
Estonia played 4 rounds from 2026 under the tree search (mcts) method, with stochastic events, 2 times. Its capability index moved from 11.49 to a median of 11.71 (+0.22), with 90 per cent of simulations between 11.57 and 11.85 [experiment_objectives].
Starting position
In 2026 the nation ranks 33 on the index against the frozen field, holds 0 of nine domains at Tier 1 and has a tier rank of 27. Its base income is 347 credits a year, calibrated so that its 2026 position is affordable with headroom for investment.
What was decided
A simulation spent a median of 350 credits on actions. The five groups that took most were Space Technologies (partner, 52 credits, 14.9 per cent); Semiconductors & Advanced Computing (buy, 44 credits, 12.5 per cent); Energy Technologies (buy, 35 credits, 10.0 per cent); Economic Output & Productivity (buy, 35 credits, 10.0 per cent); Semiconductors & Advanced Computing (build, 35 credits, 5.0 per cent) [experiment_decisions_by_group]. By mode: buy 59 per cent, partner 32 per cent, build 9 per cent [experiment_decisions_by_group]. Concentration of spending (HHI) had a median of 0.127, and the most repeated action appeared in 75 per cent of rounds.
The leading domain by period: rounds 1 to 5, Critical Technology [experiment_spend_by_period].
The method's chosen action was within 5 per cent of its best alternative in 92 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 0 shocks and 0 upside events. The most damaging per firing were S06 (1 firings in 1 simulations, −23.6 points each) [experiment_shocks].
The groups that lost most to shocks, per simulation: Trade Facilitation & Logistics −0.41; Diplomatic Statecraft & Global Networks −0.36; Monetary & Currency Capability −0.35; Fiscal Capacity & Public Finance −0.34 [experiment_shock_damage_by_group].
The edges that carried most change through the graph, per simulation: Semiconductors & Advanced Computing to Artificial Intelligence (+0.85 points); Semiconductors & Advanced Computing to Economic Leverage & Dependence (+0.35 points); Artificial Intelligence to Robotics & Autonomous Systems (+0.30 points) [experiment_propagation_by_edge].
Where it ended
At 2030 the median rank is 31 (5th to 95th percentile 30 to 32), the median count of Tier 1 domains 0 (0 to 0) and the median tier rank 28; the resilient score, the mean score sum over simulations, is 837 [experiment_objectives].
Why
Without shocks the index ends 0.000 higher on average, so the shocks cost about 0.000 index points [experiment_probe_attribution].
Holding instead of following this method changes the index by −0.376: 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 −0.019 and −0.002, for example 2030 partner Nuclear Deterrence (18 credits) [experiment_probe_attribution]. The largest actions by credits come late, when income is highest, and are rarely the most valuable.
The replay probe reproduces the simulations exactly (index change 0.000000), so the attributions are clean [experiment_probe_attribution].
Surprises
same_action (warn): the most repeated action (group and mode) appears in a median 75% of rounds.
The simulation furthest from the median (seed 1) ends +0.16 from it, with 0 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), 60 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 stochastic scenario.
The committed decision most often chosen, round by round: round 1 partner Fiscal Capacity & Public Finance (100 per cent of simulations, visit share 22 per cent); round 2 build Semiconductors & Advanced Computing (50 per cent of simulations, visit share 22 per cent); round 3 build Semiconductors & Advanced Computing (50 per cent of simulations, visit share 25 per cent); round 4 partner Semiconductors & Advanced Computing (50 per cent of simulations, visit share 22 per cent) [mcts_decision_sequence]. The search hesitated (top two root candidates within 10 per cent) in 62 per cent of decisions, considered a mean of 26 candidates, grew trees of a mean 40 nodes to depth 2.5, and took 0.0 seconds a decision [mcts_decision_sequence].
Where the committed actions came from, by proposing generator (committed simulation-rounds): marginal return + graph leverage 1.5; graph leverage 1.0; goal gap 0.5; greedy 0.5; marginal return + graph leverage | marginal return + graph leverage + goal gap 0.5 [mcts_root_candidates].
Against the greedy experiment with the same seeds (dev-est-greedy-marginal), the search's index at the horizon is −0.038 on average (10th to 90th percentile −0.053 to −0.023), better in 0 per cent of seeds; on the index −0.038, the score sum −2.8 and Tier 1 domains 0.00 [mcts_regret].
Across seeds the same root decision won in a mean 62 per cent of simulations (first five rounds 62 per cent) [mcts_stability].
The principal variation was followed 100 per cent at depth 1, 0 per cent at depth 2, 0 per cent at depth 3, 0 per cent at depth 4 [mcts_pv_vs_path].