Braess's Paradox
Learn it → Apply it → Save as Skill
Braess's paradox is the mathematically proven result that adding capacity, a shortcut, or a new option to a shared network can make the whole system worse, not better, because every independent user keeps choosing whatever looks best for them individually. It was proven in 1968 by the German mathematician Dietrich Braess, working on traffic-flow models, and unlike many mental models on this list it is not a psychological bias at all. Nobody using the new option is behaving irrationally. The paradox is a structural property of certain networks: the point where everyone's individually rational choice lands, called the Nash equilibrium, can be mathematically worse for everyone than the point the system would reach if someone could coordinate all the choices at once.
Dr. Wallace's Favorite Mental Models turns one well-chosen way of thinking at a time into something you can actually use. Each entry gives you a fast overview, an interactive walkthrough, varied real and practical examples, and an AI-ready Markdown skill you can keep.
Instead of only reading about a mental model, you can build your own AI skills library over time: a personal set of portable reasoning tools that help ChatGPT, Claude, and other AI assistants apply the thinking you value.
Braess's paradox shows up anywhere independent users share a network and each one is free to pick their own route through it: drivers on roads, requests on a server, orders in a fulfillment network, players running a set play. When a new link, lane, server, or shortcut is added, some users individually benefit from switching to it, so they do. But if enough people make that same individually sensible switch, the resulting congestion on the new shared path can raise the total cost for the whole system above what it was before the option even existed. No one chose badly. The network's shape did the damage.
Important: this is the standard textbook version of Dietrich Braess's original 1968 network, not a survey or an invented analogy. Every driver's travel time gets worse by 15 minutes after a road is added that cost nothing to use, because each driver is still, individually, choosing the fastest-looking route available at that moment.
Use the two axes below, whether a model is rooted in network routing and equilibrium or in resource use and depletion, and whether the mechanism is a provable structural property or a behavioral or systemic tendency, to see how Braess's paradox relates to three commonly confused neighbors. Select a dot or a chip to read how each one compares.
Each stage builds on the one before it. Use the numbered buttons or Next to move through the walkthrough.
Open an example to see Braess's paradox in a specific context. The first six are documented research, historical events, or published studies, the rest are common hypothetical scenarios used to teach the method, clearly labeled as such.
German mathematician Dietrich Braess, then at the Ruhr University, published a 1968 paper analyzing traffic-flow networks and proved that in certain network shapes, adding a new road, one that costs nothing to use, can raise the equilibrium travel time for every driver in the network, not lower it. The result holds purely from drivers individually choosing their fastest-looking route each time; no driver needs to behave irrationally for the paradox to occur.
Following investment in Stuttgart's road network in 1969, traffic conditions did not improve as planned. Conditions only got better after a newly built section of road was closed to traffic again, a real-world instance of removing network capacity improving overall flow, matching Braess's prediction in reverse.
In 1990, New York City temporarily closed 42nd Street in Manhattan for Earth Day. Traffic planners expected worse congestion on the surrounding streets. Instead, the closure measurably reduced congestion in the area, another real demonstration that removing a route from a congested network does not automatically make the network worse, and can make it better.
In 2003, Seoul demolished the elevated Cheonggye Expressway, a six-lane highway, as part of a project to restore the buried Cheonggyecheon stream beneath it. Rather than worsening regional gridlock, average travel times in the surrounding area improved despite the removal of a major roadway, alongside large gains in air quality and local biodiversity.
Dirk Witthaut and Marc Timme, publishing through the Max Planck Institute for Dynamics and Self-Organization in the journal New Journal of Physics in 2012, showed computationally that adding a new transmission line to a decentralized power grid can push the grid into desynchronization and outage, a direct analog of Braess's paradox in physical power networks rather than road traffic.
After New York Knicks star Patrick Ewing tore his Achilles tendon during the 1999 playoffs, the team won three of its next four games, a pattern sportswriter Bill Simmons later named the "Ewing Theory." Physicist Brian Skinner published a network-theory analysis, in the journal Significance, modeling a five-player set play as a routing network and found the team's scoring probability rose from one third to three eighths once the star player's option was removed from the model, the same selfish-routing mechanism behind Braess's paradox playing out in an offensive play rather than a road network.
An engineering team adds an expedited merge queue so urgent fixes can skip the normal review line. Within weeks, most engineers start labeling their own changes "urgent" to use the faster queue, which becomes more congested than the original single queue ever was, while the regular queue sits underused. Average time to ship, across all changes, gets worse than before the fast lane existed.
An online retailer opens a new regional "quick ship" fulfillment option alongside its two existing warehouses. Customers and the order-routing system both favor it because it looks fastest in isolation, until its limited dock capacity becomes the new bottleneck for the whole network, and average delivery time across all three warehouses ends up higher than it was with just the original two.
A city builds a new bridge directly connecting two already-busy neighborhoods, expecting shorter commutes for residents of both. Drivers from a third, previously uninvolved neighborhood start routing through the new bridge because it looks faster on a map, overloading an intersection that was never designed for that volume, and the city's average commute time across all three neighborhoods rises rather than falls.
This is an illustrative, invented scenario designed to teach the method. The figures are the agency's own working model, not documented research statistics.
Expand each item for the research and reasoning behind the model.
Yes, it is a genuine mathematical result. German mathematician Dietrich Braess proved it in a 1968 paper on traffic-flow networks, showing that for certain network shapes and cost functions, adding a zero-cost link can raise the Nash equilibrium travel time for every user. This is not a statistical tendency observed across many cases the way a cognitive bias is; it is a property that can be calculated in advance for a specific network before anything is built.
No. Most added capacity genuinely helps, which is exactly why the paradox is worth naming: it is the counterintuitive exception, not the rule. The paradox depends on a specific structural condition, roughly, that the new link lets enough selfish users shift onto a shared path whose congestion cost rises faster than the time it saves them individually. Traffic engineers can and do calculate in advance whether a proposed link is Braess-prone before approving it.
The Cheonggyecheon project removed a highway for stream restoration and public space reasons first, not as a deliberate Braess's paradox experiment, and the surrounding changes, new bus lanes, demand shifts, and other transit investment, mean traffic outcomes have more than one cause. What is well documented is the headline counterintuitive result itself: travel times in the area did not get consistently worse despite losing six lanes of highway capacity, which is the specific claim this model is used to illustrate, not a claim that highway removal alone explains every downstream effect.
No. Witthaut and Timme's 2012 study applied the same underlying mathematics to power transmission networks and found that adding a line between decentralized power generators could desynchronize the grid rather than stabilize it. The same selfish-routing logic has also been applied to computer networks, data routing, and, in Brian Skinner's published analysis, a five-player basketball set play modeled as a routing network.
The Law of Unintended Consequences is a broad description for any action that produces effects beyond what was intended, for almost any reason: incomplete information, complexity, perverse incentives, simple bad luck. Braess's paradox is a specific, named, mathematically provable mechanism within that broader territory: it only requires a shared network, independent actors each optimizing their own route, and a particular structural relationship between the routes, nothing more exotic than that.
Braess's 1968 proof shows this is false for certain network shapes. The only way to know is to model the new equilibrium after people reroute, not just the new option by itself.
Jevons Paradox needs an efficiency gain that lowers cost and raises total demand for a resource. Braess's paradox needs neither; the same number of users can produce a worse outcome purely from how they reroute.
Tragedy of the Commons is about a shared resource being depleted by overuse. Braess's paradox is about routing and congestion in a network; nothing needs to be depleted for it to occur.
Plenty of things get worse after a change for mundane reasons, a bad contractor, a budget cut, an unrelated failure. Reach for this model only when selfish, decentralized routing choices are actually driving the result.
Sometimes it is, as in Stuttgart and Seoul. Often a better fix is pricing, tolling, or coordinating the choice rather than deleting it, since the new option can still have real value for some users.
The paradox can occur even when every individual decision, by every driver, engineer, or shopper, is perfectly rational. The failure is structural, in the network's shape, not a failure of judgment by any one person.
The same mechanism has been documented in power grids, applied to computer networks, and modeled in a basketball play. Any system where independent users choose a shared path is a candidate.
if users have only one path through the system, there is nothing for selfish rerouting to reshuffle, and the paradox's mechanism cannot operate at all.
a shared fishery, pasture, or aquifer being overused and exhausted is Tragedy of the Commons, a different mechanism with a different fix, usually regulation or ownership, not routing design.
if total use of a resource rose because an efficiency gain made it cheaper to use, that is Jevons Paradox, not a routing effect.
most added capacity genuinely helps; the paradox describes a specific, checkable exception, not a general case against infrastructure, software options, or new choices.
This browser tool structures your thinking. It does not send your information anywhere or perform AI analysis.
Save as Skill converts the model into portable Markdown instructions that teach an AI how to spot a capacity or shortcut decision that could backfire through selfish rerouting, ask the right follow-up questions, and avoid the common mistakes above.
Save the skills you find useful and organize them into your own reusable library. The Markdown is intentionally model-neutral, so it can be used wherever an AI system accepts uploaded files, project knowledge, custom instructions, or reusable text context. Platform capabilities differ, so this is a portable skill file rather than a claim of one-click installation.
Describe a specific capacity, shortcut, or new option you are weighing right now. This builds a separate, personalized prompt that asks an AI to apply Braess's paradox directly to your situation, distinct from the prompt builder above.
No. Most networks are not Braess-prone, and most added capacity genuinely helps. The useful habit is checking, with a model that accounts for selfish rerouting, whether a specific proposed addition is one of the exceptions, rather than assuming either that more capacity always helps or that it never does.
Nobody's, in the sense of a mistake. The gap between what is individually best and what is collectively best, sometimes called the price of anarchy, is a property of the network's structure and the rules governing how people choose within it. The practical lesson is that the fix usually has to change the structure or the incentives, through tolling, coordination, or redesign, not through asking individuals to choose differently against their own interest.
Before adding a shortcut, a fast lane, extra capacity, or a new option to any system where people or processes get to pick their own path, model what happens once everyone reacts to the new option, not just what the new option looks like added on top of today's behavior. That one check would have caught Stuttgart's 1969 road before it opened.