Magnolia has three bridges. Seattle studied them as one.

Not long ago, I sat in a traffic jam trying to leave Magnolia. A crash had closed one of the three bridges out of my neighborhood. Roads that are usually fine just stopped.

Nassim Taleb has a famous warning: don't cross a river just because it is four feet deep on average. You can still drown in the deep part. He also points out that people tend to think the world works in straight lines, when a lot of it doesn't. Traffic is a good example. Most days are fine. The bad days are the ones that matter, and an average hides them.

That is what this post is about.

I don't like NIMBYs

For years, Seattle let homeowners block new apartments. That is a big reason rent here is so high. I want more housing in Seattle. That includes my own neighborhood.

But I dislike bad planning even more. Bad planning can turn a good idea into a bad result. And a bad result gives NIMBYs their best argument. If Seattle adds homes to Magnolia and the bridges jam up, people will not blame the bridges. They will blame the new homes. Then the next plan for more housing, anywhere in the city, gets harder.

So I built a simulator. You can try it here: Magnolia Crossing Simulator.

Three ways in, three ways out

Magnolia is a peninsula. About 22,000 people live there. You can only drive in or out on three bridges over the railroad tracks:

• the Magnolia Bridge
• the W Dravus Street bridge
• the W Emerson Place bridge

There is no other road. Every car that leaves Magnolia uses one of these three.

And almost everyone leaves. Magnolia works like a suburb. There are few jobs here, so people drive into the city for work and most other things. Today about 60,900 car trips cross the bridges each day. That is about 6 crossings for every household (some of those trips are visitors).

The Magnolia Bridge is almost 100 years old. The 2001 Nisqually earthquake damaged it. The city says it can't pay the $350 to $400 million to replace it, so it is fixing the old bridge instead.

What is coming

Three changes are planned for Magnolia over the next 20 years:

• Fort Lawton: about 500 new homes near Discovery Park. That is 45 townhomes, 355 apartments and condos, and 100 homes for seniors and veterans who were homeless. The city is still studying it.
• The new "neighborhood center": this is part of Seattle's big growth plan, the One Seattle Plan. The city's goal is 595 new homes near Magnolia Village and 312 in North Magnolia.
• More homes on each lot: new city rules let owners build several homes where there used to be one. The city does not give a number for Magnolia. I estimated one from the city's numbers for the whole city: about 330 to 885 homes.

Added up, that is about 1,750 to 2,300 new homes by 2044. That means about 3,800 to 5,000 more people. Magnolia would grow by 17 to 22 percent.

I am fine with that. I am not fine with how the city studied it.

The city looked at the average

First, credit to the city. Its traffic model knows that traffic gets worse faster and faster as roads fill up. But the growth study has two blind spots.

The first is what it looks at. The city's study treats Magnolia's three bridges as one number. It compares Magnolia in 2044 with the plan and without the plan. In the evening, the roads go from 58% full to 64% full going one way, and from 62% to 70% going the other way. That looks fine.

But this is Taleb's river. Say one bridge is 120% full and another is 40% full. The average is 80%. That looks fine too, but one bridge is jammed.

I could not find results for each bridge on its own. I could not find any results for what happens when a bridge is closed, even though bridges close for crashes and repairs all the time.

Two things from this year also matter:

• The state says the city skipped part of the study. In September 2026, a state board found the city only studied 120,000 new homes. Its zoning allows 162,847. That leaves about 42,000 homes that were never studied. The city now has one year to fix this.
• The Fort Lawton study already finds problems. It expects 2,266 new car trips a day, and it does not assume anyone takes the bus. It says the stop sign at W Emerson Place and Gilman Avenue W will go from a D grade to an F grade in both rush hours. The ramps at W Nickerson Street and 15th Avenue W also get an F. And that is just one project.

Why traffic does not grow in a straight line

Think of a road that is 60% full. Add one more car and it finds a gap. Now think of a road that is 95% full. Add one more car and it has to wait behind the car in front. Everyone behind it waits a bit longer too.

So delay does not just go up with traffic. It goes up faster and faster as the road gets close to full.

The worst of it happens at single intersections. A line of cars at one light grows until it reaches the light behind it. Now that light can't move its cars either, even when it turns green. A left turn waits for a gap in oncoming traffic that never comes. An intersection that should move one car in two starts moving one car in ten, just because the one in front of it is full. One jam spreads into the next.

This is the second blind spot in the city's growth study. It works at a zoomed-out level. It treats each stretch of road as one piece and guesses the delay from how full that piece is. It does not follow real lines of cars from one light to the next. So it can't see a jam at one light spilling into another. My simulator does track lines of cars, and lets them back up into the streets behind them.

Here is what my simulator shows on a normal day, with all three bridges open. The table shows what happens when new trips are added on top of today's 60,900.

Four things stand out:

1. Twice the traffic means about four times the delay. Going from 4,000 to 8,000 new trips makes delay 3.6 times bigger. Going from 8,000 to 16,000 makes it 3.9 times bigger.
2. Then there is a cliff. Going from 24,000 to 48,000 new trips doubles the traffic, but delay gets 28 times bigger. The roads stop clearing. At 2 a.m., 66,000 trips still have not finished.
3. A one-hour closure changes everything. This is the jam I sat in. If the Magnolia Bridge closes for one hour at evening rush, 4,000 new trips cause 4,558 extra hours of delay instead of 960. That is almost 5 times worse.
4. A long repair is worse still. With the Magnolia Bridge closed all day, the same 4,000 new trips cause 15,745 extra hours. That is 16 times worse. The other two bridges are already close to full.

So where does the real plan land? The Fort Lawton study uses about 4.5 car trips per home each day. That is less than the 6 bridge crossings per household Magnolia has today. For 1,750 to 2,300 homes, 4.5 trips each is about 7,800 to 10,300 new trips a day. On a normal day, that means 3,500 to 6,000 extra hours stuck in traffic, every day. On a day a bridge is closed, it is much worse.

The simulator starts at 1,000 homes where everyone drives. That comes out to 8,000 new trips. It is the same number you get from the city's own home goals and trip rate.

What this simulator is and is not

This is a "what if" tool. It is not a promise about the future.

• It runs 143 full-day simulations. There are 13 traffic cases (all bridges open, or one bridge closed all day or for one hour) and 11 levels of new driving.
• Traffic lights and turns are simplified. They are not the city's real light timings. The simulator also does not include traffic just passing through on 15th Avenue W.
• Each case was run once. Some results are noisy. In a few Dravus cases, adding traffic made the delay go down. That is almost surely a quirk of the simulator, not a real effect. I left it visible on purpose.
• It has not yet been checked against real traffic counts from a single day.

You can download everything: all the results, how the model works, and a review of how the city did its traffic study. If you work in traffic, please try to break it.

How I built it

I didn't write the code. An AI agent called Muse did. I wrote the prompts and made the calls. My one hands-on job was hosting: I set up a service on Railway to get the site onto my own domain.

Muse's first move was to build its own traffic simulator from scratch. I pushed it to use SUMO, a well-known open source traffic simulator on GitHub, instead of reinventing one. The method notes on the site say which engine produced which results.

The bigger problem: Muse kept taking the city's numbers at face value. It would pull the city's trip forecasts and bus-ridership guesses and treat them as facts. I had to keep telling it to stop. The version on the site now builds traffic from the ground up: homes, people per home, trips per person, and how many of those trips are by car. The city's forecasts are not used as inputs.

It took six versions to get here.

Some facts about the build:

• The traffic engine moves time forward one second at a time. Each car takes up about 7.5 meters of road. Lines of cars can back up into the street behind them. Drivers can pick a new route every five minutes.
• The street map comes from OpenStreetMap. It also uses 7,392 traffic counts from the city, each covering 15 minutes.
• All 143 runs were done ahead of time. They come to 83.5 MB of data, which your browser unpacks as you go. There is no server doing the math.

What I want

I am not asking the city to stop building. I am asking it to plan as if the homes will really get built, and as if bridges really do close.

1. Show results for each bridge, including when each one is closed. One average number is not enough for a place with three ways out. Use a model that follows real lines of cars through each intersection.
2. Study the high end for Magnolia. The state already told the city to study the 42,000 homes it skipped. Magnolia is a good place to start.
3. Fix the bridges and the buses on the same schedule as the housing. That means real money for the Magnolia Bridge. It also means buses people will actually ride. Many of my neighbors will not take the bus because they don't feel safe. You can argue about whether that is fair. But a plan that counts on bus riders it has not earned will miss.
4. Then build the housing.

If you want more housing in Seattle, you should want this too. The fastest way to kill the next housing plan is to let this one make getting out of Magnolia much worse.

Try the simulator and tell me where I'm wrong: yourfriendlynephew.com/magnolia-crossing-simulator

Sources: Fort Lawton traffic study · One Seattle Plan environmental study, Ch. 2 · State board ruling (The Urbanist, Sep 2026) · Fort Lawton status (The Urbanist, Aug 2026) · Magnolia Bridge cost (Seattle Times) · Census Reporter, 98199 · Method and review of the city's study