Closing a road can make everyone's journey shorter
In a congested network where each traveller picks their own best route, adding a link can raise everyone's travel time, and removing one can lower it. In 2003 Seoul demolished a six-lane elevated motorway carrying 160,000 vehicles a day and traffic did not get worse.
Dietrich Braess proved the effect in 1968. Selfish equilibrium and system optimum are different points, and a new link can shift the equilibrium to a worse one for every single driver — nobody defects, because each is individually doing the right thing. The Cheonggyecheon expressway removal is the best-known field case; the 1990 closure of 42nd Street in New York is another.
It is the clearest ordinary example of individually rational choices producing a collectively worse result, and the rare one where the intervention is subtraction. Most systems thinking is about what to add.
It is bounded. For networks with linear congestion costs, the price of anarchy is at most 4/3: selfish routing is never worse than a third off the optimum. That bound is oddly reassuring and also the reason the paradox stays a curiosity rather than a crisis — it is real, it is not usually catastrophic.
The same structure appears in electrical grids, in packet networks, and — most vividly — in a mechanical demonstration where two springs and two ropes hold a weight, and cutting one of the ropes makes the weight rise. Nothing about the paradox is specific to traffic. It is a property of shared-cost networks with individual routing.
Seoul was not a controlled experiment.
The expressway removal coincided with new bus rapid transit, changed signal timing and a large public campaign. The defensible claim is that removing major capacity did not produce the predicted collapse, which is still surprising, rather than that removal improved flow by itself.
A well-instrumented removal with transit and pricing held constant, showing travel times rise as classical capacity models predict.