Four views of a model whose central estimate rests on a single month

Each point is one month of the panel. The line is re-estimated when the five months of the Hormuz closure are removed, and the statistics beside it update accordingly. Removing all five reverses the sign, and removing March 2026 on its own is enough to take the coefficient below conventional significance, which suggests the estimate rests on a single observation rather than five.

Each event appears twice: the figure quoted at the time, and the change in world supply measured afterwards. Three of the six carried no barrel figure at all, a different statement from a claim of zero, and drawn accordingly.

Two panels on a shared time axis, rather than two scales on one panel. Spare production capacity is the buffer that absorbs a disruption before it reaches the price, and it approaches zero at roughly the point where a price response appears. The pattern is visible in the series themselves and does not depend on the model.

The control sets the barrels removed. The model adds what is already offline, roughly 2.67 mb/d in normal conditions, and compares the implied total against the range where observations exist. Between 5.03 and 10.05 mb/d it returns no estimate, since eleven years of monthly data contain no month in that interval. The gap is reported rather than interpolated.

Countries are grouped by the kind of exposure rather than its size. Producers inside the Gulf lost volume, since barrels that cannot be shipped are shut in at the wellhead. Import-dependent economies face a price they have no domestic production to escape. Exporters outside the Gulf kept their output and sold it into a higher price. Vertical position is the change in production between February and April 2026; horizontal position is the share of consumption met by imports.

Routes follow the shipping-lane network rather than the shortest line between two ports. The difference matters for the argument rather than the appearance: a straight line from the Gulf to Rotterdam crosses Saudi Arabia and Africa, so a route described as passing Hormuz did not go near it. A Gulf to Rotterdam route now runs 11,941 km through Hormuz, the Red Sea and Suez, and a Gulf to Ningbo route 10,341 km through Hormuz and Malacca, because the lanes go there.

The published lane data is drawn in disconnected pieces, so three corridors were completed by hand: the Red Sea between Bab el-Mandeb and Suez, the Arabian Sea between the Gulf of Aden and Hormuz, and the Cape of Good Hope. Panama and Suez are joined across the canals themselves, and nothing was connected across the antimeridian, which had been sending Pacific routes round Africa. Eleven segments are drawn rather than sourced and are recorded as such. Every route on the globe, including the country bundles and the United States import layer, now follows the same network, and the build fails if any route would fall back to a straight line. The lane network itself is drawn faint as geography, and the lanes button cycles it into a traffic view, where every segment is weighted by how many of the drawn routes travel it and any segment can be selected for its count. The lane data carries no names or traffic of its own, so that count is the one thing a segment can honestly show.

Approximating those corridors is a judgement about where ships go, which is settled geography, and it changes no measurement: every volume, transit count and barrel figure still comes from its source. Nine reference routes are checked against published sailing distances on every build, so a network that goes somewhere a ship would not fails before the page is written rather than after.

A globe rather than a flat projection, because routes between the Gulf and Asia cross the frame edge on any rectangular map and stop reading as single journeys. Dragging rotates it, the wheel or the buttons zoom, and a double-click returns to the opening view. Selecting a port opens its throughput and its country's position in the panel alongside. Chokepoints and routes are selectable in the same way, and Escape clears the selection.

Play runs the months in sequence, resizing ports, redrawing chokepoint rings and moving the Brent line underneath together, so volume and price advance under one control. Each chokepoint is ringed against its own 2024 to 2025 baseline rather than against Hormuz, which lets an unaffected chokepoint read as unaffected instead of as small. Selecting one gives its transits, its own baseline, where it ranks among the 27 this month, and the major chokepoint running closest to normal, which is the comparison that turns a large percentage into evidence. Country bundles collapse the port-to-port links between the same two countries into one arc whose width shows how many connections it aggregates.

Two shadings are available. Import dependence covers 44 countries and describes a structural position that barely moves. Stock cover, in days of net imports, covers only the six countries that both report stocks and import more than they produce, and it is the one that depletes as the months advance. Countries outside those sets are left unshaded because their position is unknown, which is not the same as being zero.

The United States import layer is different in kind from everything else drawn here. Those are measured barrels attributed to the country they came from, whereas the connection arcs record that two ports are linked by observed shipping and say nothing about what moved along them. Marker density on the arcs divides an origin's measured loadings evenly across its connections, which is an assumption stated wherever a number derived from it appears.