It wasn't the weather

One text from an airport gate, one rebuilt day of US air traffic — all 21,261 flights of it.

It started with a text. Thomas, stuck at a gate on June 12, delayed an hour, "because of weather":

There's an incredible amount of almost incomprehensible variability. How can you create a system that is priced that meets that

Incomprehensible is a dare. So Eric rebuilt that exact day of the US air-traffic system — all 21,261 scheduled domestic flights — as a simulation that runs in your browser, with sliders for the things people blame. Then Thomas watched it cold, on camera.

What the cause data actually says

In the federal delay-cause data for that day, late-arriving aircraft — the delay your plane inherited from wherever it flew before it got to you — accounts for 40.3% of all delay minutes. Weather gets 7.9%. Five to one. Your "one-hour weather delay" is usually a morning delay somewhere else, compounding its way across the country until it lands on your gate. That mechanism — delay as something planes carry, not something skies do — is the whole episode.

Is my flight delay really because of weather?

Probably not in the way the gate agent means it. When a US flight runs late, the airline files the delay minutes into five federal cause buckets: carrier, extreme weather, National Aviation System, security, and late-arriving aircraft. The weather bucket only counts weather severe enough to be coded as such; an ordinary thunderstorm slowdown mostly lands under "National Aviation System" as capacity restrictions. And here's the trick that makes "weather" feel true even when the sky above you is blue: yesterday's storm in Dallas shows up today as a late-arriving aircraft in Boston, three flights downstream. Weather starts cascades. But the thing that actually eats your afternoon is the cascade itself — across all of 2025 (7,001,619 flights), late-arriving aircraft accounts for 39.2% of delay minutes and extreme weather 6.4%, nearly the same five-to-one ratio as Thomas's day at the gate.

How one late plane becomes a national evening

The simulation propagates delay the way the real system does, three ways at once. Down each aircraft's rotation: the federal record includes tail numbers, so we know your 5 p.m. plane is the same metal that flew three legs since sunrise, carrying whatever it collected. Through airport capacity: when weather cuts an airport's arrival rate, a queue forms, and the queue delays every airline equally — that's how one storm becomes a cross-airline mess. And into cancellations, via a crew duty-day proxy: push a rotation late enough and the last flight doesn't go at all. Replay June 12 with the sliders untouched and the model reproduces the day; nudge morning turnarounds a few minutes tighter and you can watch a clean map melt by dinner.

Can you price incomprehensible variability?

That was Thomas's actual question, and it has a real answer: the variability isn't incomprehensible, it's the predictable output of a system that chooses to run hot. Planes, gates, and crews are expensive to leave idle, so airlines schedule close to capacity — and basic queueing math says waiting time doesn't grow gently as utilization approaches 100%, it diverges. The industry prices itself right up against that cliff because most days it works. Weather doesn't create the fragility; it just supplies the push. That's why the same one-hour storm costs you ten minutes at 8 a.m. and three hours at 6 p.m., and why the episode's verdict line stuck: the system doesn't break, it melts.

Break it yourself

The simulation is public and interactive: paint your own storms onto the map with God Weather, or squeeze turnaround times and watch one late morning cascade across the country. If you can make the model say Thomas was right, we want to hear about it.

Episode post · published 2026-07-18. The verdict on the scoreboard: Math +1. Thomas has filed a formal protest, which we have framed.