The frequency that went quiet

For one week every July, a field in Wisconsin runs the densest arrival stream on earth — and it does it by giving up the two things every other airport is built on: your callsign, and your readback. We pulled every position report and measured what that actually produces.

The FAA publishes a notice for this. It is worth reading the sentence that governs the whole operation, verbatim:

“At Fisk Controllers will call your aircraft by color and type (if known). No verbal responses are required.”

No callsign. No readback. No confirmation that you heard anything at all — you rock your wings and you land. Every other tower in the country is built on positive two‑way radio, and this one, for one week a year, throws it out.

What it produces

On Sat 18 Jul, the Saturday before the show opened, we detected 854 touchdowns at Wittman Regional. In the busiest hour there were 106 of them — one every 34.0 seconds, sustained for sixty minutes. In the busiest ten minutes it tightened to one every 22.2 seconds.

At the peak, 42 aircraft were on the arrival corridor simultaneously, strung out single file over a railway line. And 597 of the day’s 853 gaps between one landing and the next were under a minute.

Nobody told the computer the procedure

This is the part that made the video worth making.

The pipeline was never given the arrival procedure. It was given raw position reports and asked three questions: which way is the traffic pointing near Ripon, what altitude is it flying, and how fast is it going. Here is what came back, against what the FAA notice prints.

* The two bearings and the two altitudes are the same number. The speeds are not quite the same quantity: the notice sets 90 kt indicated airspeed, and ADS-B broadcasts groundspeed, which the wind moves. A 92 kt median ground against a 90 kt indicated target is a close match, not an exact one, and it is reported that way here rather than rounded into agreement.

Hundreds of strangers, in everything from a Cub to a Bonanza, with nobody confirming a single instruction, flew the published profile closely enough that you can recover the procedure from the raw tracks. That is not a coincidence you can fake. It is what discipline looks like when you measure it from orbit.

A second, unplanned check fell out of the same data. The notice closes the field to arrivals between 8 PM and 7 AM. We never encoded that. Of 4,503 detected touchdowns across the week, just 35 — 0.78% — fall outside it, and on the peak Saturday the figure is zero. A handful of late arrivals is exactly what you would expect from a real field with real exceptions; a clean zero across nine days would have been a reason to distrust the detector.

The control

A measurement that only works at one airport proves nothing, so the identical detector — same code, same thresholds, same definition of a touchdown — was pointed at O’Hare on the same days.

The fair comparison is not a single runway’s rate. Assigning a landing to one of several closely‑spaced parallel runways is the least certain thing this detector does, and O’Hare’s parallels are a few hundred metres apart. So the statistic is arrivals per runway in use during one hour, which survives a per‑landing mis‑assignment.

O’Hare is separating heavy jets under instrument rules with wake‑turbulence minima, radar, discrete transponder codes and a readback for every instruction. Oshkosh is a controller standing in a field at Fisk, calling aircraft by their paint.

What we cannot see, and which way it bends

Aircraft originally certificated without an engine‑driven electrical system are exempt from ADS-B Out — and AirVenture is exactly where those aircraft are. The antiques, the ultralights, the fabric taildraggers: invisible to this dataset.

That bias only runs one way, which is the useful thing about it. The aircraft we cannot see landed between the ones we can. Every gap reported here is therefore too long and every rate too low. This is a floor.

The beat that got cut

The first draft of the film had a fourth act arguing that the readback was abandoned because the frequency could not physically carry it. It is a tidy story and the data does not support it.

At the measured rate aircraft crossed Fisk — a peak of 90 an hour, one every 40.0 seconds — sequencing calls would occupy a single shared frequency roughly 18–28% of the time with readbacks included, against 10–15% without. That is a busy channel, not a saturated one.

So the claim was cut rather than dressed up. The radio timings are the one modelled input anywhere in this project, and they are reported as a band for exactly that reason.

A number we threw away

A figure of “21,883 operations, 11 days, 20–30 July, about 148 an hour” circulates in search results attributed to AirVenture 2026. It is not in any EAA release, and it cannot be a 2026 figure — a window ending on 30 July could not have been reported while the event had only just closed. EAA’s own published counts are 16,780 for 2024 and 16,246 for 2025. The figure was discarded and nothing here is built on it. There is a gate in the build script that fails if any official 2026 total is ever quoted, because none exists yet.

What actually landed

The ten most common types in the measured stream, which is as good a portrait of American general aviation as you will find anywhere:

The part that generalises

The instinct everywhere else in aviation — and in most systems people design — is that safety comes from confirmation. Say the name, read it back, prove the message arrived.

Oshkosh is the working counter‑example. For one week it strips out identity and acknowledgement entirely, replaces them with colour and eyesight, and gets the densest arrival stream in the world. It only works because everybody arrives having already read the same document and agreed to fly it exactly — which is the actual lesson, and it is not really about aeroplanes.

Sources: adsb.lol globe_history 2026 daily archives, ODbL · FAA Domestic Notice dom26020_sp, EAA AIRVENTURE OSHKOSH 2026 · ourairports runways.csv · EAA AirVenture by-the-numbers releases (ceiling check only). Every figure reproduced from the raw files by data/oshkosh_stream.py.