FY2050 · KPMG STRATEGY & VALUE CREATION

The Helios Mission

Helios Vega Interplanetary Logistics operates 15 nodes across Earth, orbit, and the lunar surface — for a baseline of €156M in annual fixed cost. The Board wants to know which to keep open under genuine uncertainty. This is your analysis dashboard.

15
Nodes
75
Customers
9
Scenarios
4
Quarters
€156M
Baseline Cost
2^15
Configurations

The data pack

Eight files describe the system. Click any card to inspect the raw rows — distances, demand distributions, scenario probabilities, cost rates.

Computing the answer
Initializing...
1
Loading network and customer data
2
Simulated Annealing · 2,500 iterations
3
Computing 9-scenario cost model
4
Monte Carlo · 1,000 trials with Gamma demand
5
Building trade-off comparisons
FY2050 · STRATEGY & VALUE CREATION · KPMG

The Helios Mission

Designing an interplanetary logistics network under genuine uncertainty. Our recommendation: operate of 15 nodes at a probability-weighted expected annual cost of . The network sustains all 9 disruption scenarios and holds under Monte Carlo stress testing.

Expected Cost
95th Percentile
CVaR 90% Tail
vs. All-Open Baseline
SCROLL TO EXPLORE

From brief to defensible network

The dashboard is not just a visualization of an answer. It documents the rules that were applied so the Board can see why the recommendation is robust: single-source quarterly flows, no flexible capacity, disruption reassignment, explicit workforce/treaty clauses, and Monte Carlo gravity-factor risk.

1. Network decision

Each candidate node is treated as an open/closed binary decision across all 15 nodes. The dashboard recomputes the open/closed set live; off-world facilities are not locked open, but any closure triggers the relevant treaty cost.

15-node binary decisionno off-world shortcut

2. Flow assignment

For every customer-quarter, demand is assigned using the briefing-card heuristic: nearest feasible open node with remaining quarterly capacity. Quarterly capacity is exactly annual capacity divided by four; if capacity is insufficient, the served part touches only one node and the remainder is unserved and penalised. This is disclosed as a greedy heuristic, not a full min-cost-flow optimiser.

single-sourcecapacity / 4unserved penalty

3. Scenario costing

All nine disruption scenarios are evaluated and probability-weighted. A disrupted node is not closed: it still pays fixed cost, but its affected-quarter capacity is zero and its customers must be reassigned.

9 scenariosS7 = Q4 × 1.30 onlyfixed cost remains

4. Workforce and treaty discipline

Workforce agreements are not hidden footnotes. The model detects every explicit DC closure clause: WR01 German dual-closure, WR02 Spanish dual-closure, WR03 Lyon retraining, WR04 Endurance deorbit, and WR05 Brand Base dormancy/repatriation. WR06 is treated as a hard guardrail rather than a cost item: terrestrial closures may not exceed six, and the financial impact remains zero.

WR04 countedWR05 countedWR06 respected

5. Uncertainty layer

After the network is fixed, Monte Carlo samples 1,000 futures from Gamma demand distributions. For flows touching distant off-world nodes, the model preserves mean demand and multiplies CV by the assigned node's demand_variance_factor before sampling. This follows the README wording: uncertainty grows with lead time, not baseline demand.

1,000 MC trialsGamma demandgravity CV inflation

6. Validation

The output includes three hand-check transport calculations and a compliance checklist. This is designed to make the model auditable: a reviewer can recompute a sample route and trace the six cost components.

hand checkssix cost componentsauditable

From Earth to Mars, a map of distances

The Helios Vega network spans four operating volumes. Cargo committed to distant nodes locks in weeks or months ahead — time dilates with distance, and uncertainty grows with it. This is the conceptual centre of the problem.

Operating Volumes
4
Customer Zones
75
Max Lead Time
10 wk
LEGEND
Terrestrial Hub
LEO Station
GEO Relay
Lunar Base
Mars Outpost

Try it yourself · play with the network

This is where the dashboard becomes a sandbox. Toggle any of the 15 nodes on or off, fire a disruption scenario, drag the cost sliders — everything recomputes live against the same engine. Hit ★ Recommended any time to restore the full SA-optimal network and explore from there.

Nodes · click to toggle

Presets

Scenarios · click to apply

Cost parameters

Monte Carlo

Monte Carlo idle · deterministic mode: same parameters + same seed = same answer.

Live results · for current selection

Open Nodes
Expected Cost
vs. Recommended
Selected Scenario Cost

Cost by component · selected scenario

Monte Carlo distribution · selected scenario

The recommended network

Recommended node configuration across terrestrial, orbital, and lunar operating volumes. Open nodes serve customers under the single-source assignment rule, by quarter.

Open Terrestrial
Closed
Off-World
Customer Zone
Assignment Flow

The six components

Fixed, transport, SLA penalty, carbon, workforce/treaty, and unserved demand. Computed across all 9 scenarios, weighted by probability. The number reconciles.

Cost by Component
Transport Cost by Quarter (Nominal)

When something goes wrong

Performance across the 9 official disruption scenarios plus two custom stress tests we designed. Only the 9 official scenarios are probability-weighted into the headline expected-cost metric; custom stress tests are shown separately and do not contaminate the recommendation number.

Cost by Scenario (EUR millions)
IDScenarioProbabilityNodes Offline QuartersCost (EUR)Unserved Units / Carbon Δ

Official rows S0–S8 are the only rows used in the probability-weighted expected-cost headline. Custom rows are separate stress tests for analytical depth and are not assigned probabilities.

1,000 futures

One thousand Monte Carlo trials with Gamma-distributed demand, scenario sampling, and the gravity factor inflating variance for off-world flows. The histogram is the full distribution. The SA-optimised probability-weighted cost shown in the hero is recomputed live in the browser from the embedded data. The stochastic MC mean may differ because it samples demand uncertainty and disruption scenarios; gravity is implemented as mean-preserving CV inflation by assigned node.

MC Mean (Stochastic)
Avg over 1,000 trials with demand uncertainty & gravity factor
Median (P50)
95th Percentile
CVaR 90% (Tail Risk)
Monte Carlo Cost Distribution — 1,000 Trials

How we solved it

Two techniques, executed and validated. Simulated Annealing to search the 32,768-option space, and Monte Carlo to stress-test the answer under genuine uncertainty.

🔥 Simulated Annealing

SA explores network configurations by making random open/closed changes. It accepts improvements, and occasionally accepts worse moves (controlled by a "temperature" that cools over time) to escape local optima.

1
Initialize: Start from a feasible greedy/random network. All 15 nodes, including off-world facilities, are eligible open/close decisions; closure is allowed but treaty cost is counted.
2
Evaluate: Compute full 6-component cost across all 9 scenarios, probability-weighted.
3
Propose: Flip one node's open/closed state randomly.
4
Accept/Reject: Accept if better. Else accept with probability exp(−Δcost / T).
5
Cool: T × 0.997 each iteration. Run 2,500 iterations. Keep the best.

T_start = 3,000,000 cooling = 0.997 iterations = 2,500

Defence: The search uses the challenge-card neighbour rule: one node is flipped per iteration. When the user clicks Run Analysis, the browser actually recomputes SA from the embedded CSV data, then recomputes the selected network across all 9 disruption scenarios and runs Monte Carlo. No hidden precomputed answer is rendered as the final result.

🎲 Monte Carlo Simulation

Once SA delivers the recommended network, MC stress-tests it 1,000 times with random demand and random disruptions, producing a full probability distribution of cost outcomes — not just a single number.

1
Fix network: Lock the SA-recommended open/closed configuration.
2
Sample demand: Preserve each customer-quarter mean, multiply CV by the assigned node's demand_variance_factor, then draw from the adjusted Gamma distribution.
3
Sample scenario: Pick one disruption scenario from the probability table each trial.
4
Compute cost: Full 6-component model. Collect 1,000 outcomes. Report mean, P95, CVaR.

trials = 1,000 Gamma: Marsaglia-Tsang gravity factor applied

Defence: MC samples demand from the supplied Gamma parameters, applies mean-preserving CV inflation using the assigned node's variance factor, samples one of the 9 official scenarios by probability, and reports mean, P95, and worst-10% tail risk. The custom stress tests are deterministic extras, not inputs to the official expected-cost metric.

SA Convergence — Best Cost Over Iterations

Alternative futures

Three network configurations explored: SA-recommended optimum, a cheaper-but-riskier minimal network, and a greener-but-pricier configuration. Plus a cost-vs-resilience frontier.

Cost vs. Tail Risk (CVaR) — Pareto Frontier
Cost vs. Carbon Emissions

What the AI cannot know

The model gives a cost-minimising answer under the rules we specified. The partner-level work is making the assumptions visible: where treaty cost is counted, where social risk remains, and what the Board should not mistake for certainty.

🧠 Why this is not a blind AI answer

The chosen network closes DC11 Prague, DC12 Lisbon, DC13 Endurance Station, and DC15 Brand Base, while keeping DC14 Cooper Relay as the off-world operating node. That answer is only defensible because the cost model explicitly counts the closure consequences rather than silently treating closures as free.

🛰️ Treaty obligations were counted, not ignored

Closing Endurance Station triggers WR04: €4.5M for deorbit/safe-disposal under the Outer Space Treaty Annex VII. Closing Brand Base triggers WR05: €2.8M for crew rotation, repatriation, and dormancy maintenance under the Artemis Operations Compact. The recommendation keeps these costs visible in the Workforce/Treaty component, which is exactly the kind of hidden constraint the brief warned teams not to miss.

⚖️ Workforce feasibility guardrail

The model enforces the Pan-European Works Council rule that terrestrial closures may not exceed six in the exercise. This recommendation closes only two terrestrial hubs, so it remains feasible. It also avoids the German dual-closure settlement and Spanish dual-closure settlement by keeping one or both relevant hubs open.

📊 Residual risk — what could still go wrong

1. Assignment is greedy, not globally exact. This follows the challenge recipe and is displayed transparently, but another assignment heuristic could shift marginal flows. 2. Treaty cost is assumed fixed. A real negotiation could make DC13/DC15 closure more political or more expensive. 3. Gravity factor captures demand drift, not operational failure. It inflates demand CV but does not model launch failure or longer-than-planned lead times. 4. Scenario dependence is simplified. The nine scenarios are probability-weighted independently; a real energy crisis, labour action, and cosmic disruption could correlate.

⏱️ What we would test with more time

Run more SA restarts, test alternative assignment heuristics, create a full cost-vs-tail-risk frontier, add multi-region concurrent disruptions, stress treaty exit costs, and validate customer relationship risks with commercial leadership. The current answer is a strong analytical baseline; the Board decision should also weigh customer, labour, and treaty judgement.