World trade atlas
Public. Click any country for its economy, ports, chokepoint dependencies and bilateral trade. Toggle the maritime network — shipping lanes, 300 ports, 28 straits.
About Autonaly
What happens to the world when physical supply breaks?
When a canal is blocked or an export ban lands, someone has to answer who gets hurt, and how badly. That answer usually takes an analyst hours of assembling trade data by hand — and most published answers are qualitative guesses. Autonaly computes it: the countries exposed, their dependency ratios, and the dollars at risk, traceable to public customs data and a published formula.
Gemini reasons and routes. A deterministic engine computes. A human approves.The model never generates a number.
How a briefing happens
- 1
A signal arrives
A wire headline lands on the event stream — a canal blocked, a strait threatened, an export ban announced. Nobody pushes a button.
- 2
The desk routes it
A coordinator classifies the event and hands it to one of three specialist agents — energy, food security, or technology supply — each carrying its own domain knowledge. Financial crises are refused: trade data cannot see them.
- 3
Severity is measured, not assumed
For a chokepoint event, the specialist reads real vessel transits from IMF PortWatch. When a headline says “halted” but the ships say 17%, the ships win. When the feed itself looks degraded, the agent refuses to score at all and escalates.
- 4
A deterministic engine computes
Exposure scores, dependency ratios and dollars at risk come from an engine with no AI in it — the same maths you can drive by hand in the simulator. The model never generates a number; a guard rejects any narrative containing a figure the engine did not produce.
- 5
A human approves
Nothing publishes without a person clicking approve. Briefings the agent could not score honestly arrive unscored, with the reason quoted verbatim.
Speculate on the next crisis
The desk reacts to events; an analyst gets paid to think ahead of them. Read the political news, pick the country you think breaks next, and the simulator builds the crisis for you: every commodity where that country supplies at least 1% of world trade becomes a disruption channel, and the engine ranks who gets hurt, by how much, in dollars. Ask the desk for a written read — grounded in a curated century of supply crises, so your speculation arrives with its historical reference class: history doesn't repeat, but it rhymes. Save the scenario to your dashboard, and take the brief with you — as Markdown, or as a real Google Doc written into your own Drive. Reopening a saved scenario replays it on the live engine — the numbers are reproducible, not remembered.
The exported document is the whole analysis, not a summary of it: key figures, the written read, and then every disruption channel as its own section — the countries it reaches, what that exposure is worth against their GDP, what the country at the centre of it stops earning, and who gains — each with its own map. It closes with the historical reference class, what the model does not see, and a glossary explaining every measure, because a score nobody can interpret is not evidence. Autonaly asks only for permission to create files, so it can write into your Drive but cannot read anything already there.
The judgment stays yours. The desk deliberately refuses to say how likely your scenario is — it computes exposure, states what the model cannot see, and leaves the probability, the timing and the conclusion to the person whose name goes on the summary.
Four surfaces
Scenario simulator
Public. Stress-test a chokepoint, a port, or a whole conflict — including a custom crisis in any of 88 countries, with disruption channels derived from the trade data rather than curated. Then ask the desk for its read.
Your analyst
Sign in and configure a personal analyst — the commodities, countries and chokepoints it watches. When an event touches your watchlist, its note is waiting before you ask. Your saved scenarios live here too.
Review queue
The human gate. Every briefing the desk files waits here for approval — including the ones it refused to score, which is the honesty that makes the rest credible.
The data, and its limits
11.25 million bilateral trade rows from CEPII BACI, 24 curated commodity baskets, live vessel transits for 28 chokepoints from IMF PortWatch, country context from the World Bank, and 88 countries with a large enough world-trade position to anchor a custom crisis scenario. Every exposure figure uses latest-year trade weights and models first-order effects only — no inventories, no substitution, no probabilities. Those limits are stated on every page that shows a number, because a reference product that hides its assumptions is not one.
Data: BACI/CEPII (Etalab 2.0) · UN Global Platform; IMF PortWatch · World Bank WDI (CC BY 4.0) · Methodology 1.0.0
Hire your own analyst
Tell it what to watch — it reads every event so you don't have to. Then speculate on the next one before it happens.