Skip to content
Methodology 2 of 8

How Monte Carlo Simulation Works

Running ten thousand futures to understand the one you are living in.

Prediction is impossible. The world is too complex, too nonlinear, and too full of genuine randomness for any model to tell you what will happen next quarter. But something more useful than prediction is possible: you can map the space of plausible futures and understand the shape of your uncertainty.

This is what Monte Carlo simulation does. Instead of asking “what will happen?” it asks “what could happen, and how likely is each possibility?” The answer is not a number. It is a distribution — a landscape of outcomes, each weighted by its plausibility.

How it works

When you run a scenario in Compass, the engine executes the same computation ten thousand times. Each run starts from the same initial conditions — the current state of the six Compass nodes, the current coupling matrix, your specified shocks and interventions. But each run adds a slightly different pattern of random noise, representing the irreducible randomness of the real world: an unexpected election result, a pipeline explosion, a viral social media moment.

The result is ten thousand distinct trajectories — ten thousand possible futures. Some converge toward stability. Others spiral into resonance. Most cluster around a central tendency, but the tails contain the scenarios that matter most: the low-probability, high-impact outcomes that define real risk.

The fan chart you see in the Scenario Lab is a compressed view of these ten thousand trajectories. The dark central band shows where 50% of futures land (the 25th to 75th percentile). The lighter bands extend to the 10th and 90th percentiles. The full extent shows the 5th to 95th. If the fan is narrow, the future is relatively predictable. If it spreads wide, you are in a regime of high uncertainty — and the width itself is information.

Why coherent seeds matter

Here is a subtlety that matters enormously and is invisible to most users: when Compass compares two scenarios — say, “Taiwan blockade” versus “negotiated resolution” — both scenarios receive exactly the same sequence of random noise. Trial 4,217 in the blockade scenario encounters the same random shocks as trial 4,217 in the negotiation scenario. The only difference between the two is the structural change you specified.

The coherent-seed guarantee: differences between scenarios are 100% structural and 0% stochastic. If the blockade scenario produces worse outcomes than the negotiation scenario, that difference is real — it is not an artifact of one scenario drawing unluckier noise.

Compass achieves this through pre-generated noise libraries called SLURPs (Stochastic Library Units with Relationships Preserved), aligned with the open SIPmath 3.0 standard.

How to interpret results

When reading simulation output, resist the temptation to focus on the median line. The median is the single least interesting part of the result. What matters is:

  • The width of the fan. Wide means uncertain. Narrow means the outcome is relatively determined by the scenario you specified.
  • The shape of the tails. Are the bad outcomes much worse than the good outcomes are good? That asymmetry is the signature of fragility.
  • Where the fan splits. If the fan suddenly widens at a particular point in time, the dynamics are passing through a tipping point where small differences in noise lead to large differences in outcome.
  • How two fans compare. When comparing scenarios, overlap means the evidence does not clearly discriminate between them. Separation means the structural difference has a measurable effect.

From the book: This methodology is developed in depth in Beyond Synthesis, Chapter 21–22: “The Navigator Factory.”