Toronto closed roads for the World Cup. Expressway traffic fell up to 30%.
Not a fluke. A 1968 math paradox most planners forgot. Here's what happened, and why no city dares to use it on purpose.
Closing a road can speed up traffic.
It is called Braess's paradox (1968). Add a road and selfish routing can make everyone slower. Remove one and the network can flow. Real-world echoes have repeated for decades, and are almost never used on purpose, because no one can predict when it will work.
Tore out an elevated freeway. The predicted gridlock never came.
Closed Broadway at Times Square. Traffic held up, injuries plunged, the plazas became permanent.
Closed stadium-area roads for the FIFA World Cup. Expressway volumes fell 10–30% on match days.
It is not engineering. It is fear.
Close a road and people switch to transit: you are a visionary. Close it and they do not: you have paralyzed the city and ended your career.
So cities never try. The one move proven to cut congestion sits unused, because the downside of guessing wrong is catastrophic and no traffic model can tell you which way it breaks.
PhD models. Months. Then a gut call.
Consultants build a traffic simulation over months.
It predicts flow at today's driver volumes.
A committee makes the call on politics and nerve.
The flaw:a model built on today's traffic predicts the average, and gets this exactly backwards. It assumes the same drivers on fewer roads, so it forecasts gridlock. It cannot see that the drivers themselves change.
Demand is not fixed. People are the variable.
Toronto closed roads for the World Cup, and the city did not seize. Citywide travel times held steady, because demand moved. An average-based model never sees this. It has no way to know a commuter will leave the car home. That behaviour is the whole answer.
Source: City of Toronto / TTC / Bike Share Toronto / Toronto Star · FIFA World Cup 2026, June match days
AI writes the plan as a model you can read.
State the goal in plain language. An LLM formulates it as an explicit optimization model. A solver returns the provably best plan. No PhD, no six months. Explicit goal, explicit limits, so a mayor or auditor can read it line by line.
“Keep the city moving. Cut downtown car volume.”
Then AI agents play your commuters.
A closure changes behaviour. History cannot tell you how. So we simulate the people, and find the tipping point, before a road is touched.
Takes transit if it is credible and close.
6 a.m. shift, kids. Drives no matter what.
Bikes when three friends already do.
The tipping pointEnough switch and the city flows. Too few and it seizes. The plan re-solves until enough people move for it to hold.
Same road closed. Known in advance.
Toronto ran the experiment for real, once, for the World Cup, and it worked. The point is not that they were wrong. It is that they could not have known first. Simulation is how you know before you close the road.
Illustrative, on public Toronto data. We did not forecast this event.
We did not predict Toronto. Nobody did. That is the point.
A city ran the gamble once and got lucky. The question we answer: how do you know before you close the road? That is the difference between a bet and a decision. Optimization for the answer. Simulation for the truth.
Have a decision like this?
We take on a small number of design partners. Bring the plan you are afraid to run.