War Room

How Simulation Works, Transparency & FAQ

Before you act on a scenario you should be able to answer one question about every number in it: where did this come from? That question is the whole reason this page exists. It opens the hood in plain language — what the engine computes, what the AI analysts add on top, and where the deliberate judgment calls sit — so that you can read each result at the level of precision it actually supports, and defend it to whoever asks.

The four moves of a scenario run

1. Turning the headline into price pressure — Each shock you stage becomes two effects on every commodity it touches: a push on the likely direction of prices, and a widening of the range of outcomes. The sizes come from expert-tuned rules of thumb scaled to real-world footprint — an armed conflict matters in proportion to the combatants' combined share of a commodity's production, an inflation surge in a major economy lifts the whole basket priced in its currency, a supply cut pushes prices up. If your scenario contains several shocks, their effects add up. This is the move that converts a headline into a concrete pressure on each price you care about, and everything after it builds on the result.
2. The producers' negotiation, on producer-quota scenarios — When a scenario contains a producer-group quota cut, the engine does not take your number at face value. It simulates the bargain between the world's two largest producers of that commodity: the bigger one moves first, knowing the rival will respond in its own interest, and the engine then adopts the supply change from the deal they would actually strike — replacing the quota figure you typed. The Stability score tells you whether that deal would hold. A high reading means the second producer gains almost nothing by quietly breaking it, so the agreement sticks; a low reading means announced cuts tend to leak away in practice. That is what stops you over-budgeting for a dramatic announcement the market's own incentives would water down.
3. The ripple — The shock then travels outward through statistically measured relationships built from years of real price data. Durable tethered pairs — prices tied together over the long run, which reliably snap back into line — carry the signal first; otherwise only meaningfully strong correlations do. The effect weakens at each step away from the source, and the ripple deliberately stops after a small number of steps. That is why the Map draws commodities on concentric rings around your shock, and it is why the map is a second-order radar rather than a list of the exposures you already knew about: it surfaces the ones you would not have thought to ask about, such as a gas disruption quietly lifting fertiliser costs.
4. Replaying the coming months, many thousands of times — For every affected commodity the engine fits a price model to that commodity's own real history — trending markets drift, mean-reverting markets get pulled back toward a long-run level — and then replays the coming months many thousands of times over: once without your shock, and once with it. Every number you see on screen is the difference between those two futures, which is why the output reads as “this commodity's most likely outcome moves by about this much, and here is the plausible worst case” rather than “prices would rise”. Commodities without enough real price history are left out rather than guessed at. Whether a market's whole mood would flip is part of the same output — see Market Regimes & Risk Analytics.

The AI analyst desk

The analysts only take over once the math is finished, and they work in the same order as the progress dots on the run page, each handing its output to the next. Their full write-ups live in the Log tab and in the Intel drawer, which is the audit trail: the place to verify the chain link by link before a result leaves the room. Every measure they refer to is defined in plain language in the analytics glossary.

Model transparency and limitations

Start with what the numbers are built on, because the foundations are real: the correlations, the tethered hedge-pair relationships, the market-regime states and every price history used in the simulation are actual market data, re-fetched and re-fitted at the moment your run starts rather than cached from last quarter. That is also why saved scenarios shared with your organisation are re-run automatically each morning after the daily data refresh — a scenario saved months ago keeps tracking the market as it is today, so a team's stress-test library never quietly goes stale.

Between that real data and the final figures sit deliberate simplifications. They are worth knowing, because they tell you how precisely to read a result — and they are the reason to read the whole output for what it is: a modelled range of outcomes for a hypothetical event, not a prediction that the event will happen, and not a promise that prices would land where the median sits.

Frequently asked questions

• Why can't I run a scenario? Staging and running scenarios has to be enabled for your organisation — your account manager can arrange it. Browsing the preset loadouts is open to every signed-in user, so you can explore the catalogue either way.
• Why is a commodity missing from my results? The engine needs enough real price history to simulate a commodity honestly, and anything with too little is left out rather than guessed at. It is a deliberate choice: every number you do see is backed by real history.
• What does the “Projection” banner mean? The full simulation was not available, so the view shows directional estimates only — which way prices lean, and roughly how much. Use it for orientation, not for budgeting.
• Why did my quota number change? The producer negotiation replaced it: the engine simulates the deal the largest producers would actually strike and adopts that supply change instead. The gap between your input and the outcome is itself the insight — announced cuts and delivered cuts are different things.
• Why is there a run I didn't start? Saved scenarios shared with your organisation are re-run automatically each morning after the data refresh. Those runs carry a Scheduled chip, appear in the event log as started by the scheduler, and cannot be cancelled.
• Why is the Game tab empty? “Awaiting GameTheory agent” means that analyst has not finished yet. One element is scenario-specific: the engine-computed negotiation and its payoff grid appear only on scenarios containing a producer-group quota cut, and other scenarios carry the analyst's strategic commentary instead.
• What does a red parse badge mean? The Macro analyst was not confident it read your typed prompt correctly. Review the proposed shock list and any parse warnings before running, and correct whatever it misread — a misread prompt produces a confident answer to the wrong question.
• Can my colleagues see my scenarios? Yes. Saved scenarios are shared across your whole organisation by default, so everyone works from the same library. The one exception is a pair-trade stress test exported from the statistical-arbitrage screen, which stays private to you because it encodes your own trade idea.
• Is this investment advice? No. The numbers come from the market-data engine, the words come from the AI analysts, and every content page of an exported report carries the not-investment-advice footer. Treat every written summary as commentary to check against the numbers it describes.
• How do I get access, or add a commodity? Contact your account manager.