Hearing a plane is easy. Naming it is not.

Catching an aircraft is the easy part. Every plane over the terrace broadcasts its position, altitude and speed, plus a radio address that belongs to that airframe for life. Our antenna hears all of it, thousands of times a day.

The hard part is a question that sounds trivial: which airline is flying it? We have three clues, and each one sometimes lies.

1. The flight's name. In flight, most aircraft broadcast a callsign, like TAP96AW for a TAP flight. It is the best clue, because it names whoever is actually operating that flight. But some planes only start broadcasting it after they have passed us, and private jets and military aircraft often use codes that name no airline at all.

2. Public registries. They tell you who owns an aircraft, not who flies it today. Airlines rent planes from each other, sometimes with the crew included: this summer, Latvian airBaltic jets flew Swiss routes to Lisbon. Registries also age badly. One Finnair jet is still listed as British Airways, after a short loan back in 2022.

3. The radio address itself. It is tempting to read it like a label, but it is allocated by country, a bit like a phone country code. An old rule of ours did exactly that, and got it wrong at scale. Every Swiss-registered plane became "Austrian Airlines". About 1,000 Azores Airlines flights were counted as TAP. Our May report even named Austrian Airlines among the quietest airlines of the month, although not a single Austrian flight had ever passed.

So, on 26 and 27 September, we rebuilt the method. The clues are now used in a strict order, and when none of them is sure, we leave the airline blank rather than guess. About 2,000 flights have been reassigned. The noise figures did not change, only the names attached to them, and each published report now carries a correction note. When a flight is operated with another airline's aircraft and crew, the gallery now says so.

Then, while preparing the September report, the microphone turned out to lie too. Now and then the sound meter records a single electrical spike: one reading, a tenth of a second long, sometimes long after the plane has gone. Real aircraft noise never looks like that; it rises and falls over several seconds. Ten September flights had been credited with peaks of up to 108.9 dB that never happened. They now carry the real level measured as the plane passed, and the original value stays on record. For June to August, the independent second-by-second log that lets us check this did not exist yet, so the peaks above about 105 dB in those reports are now flagged as unverifiable.

The September report, published today, is the first built on the cleaned data. Its loudest flight is a real one: an Emirates Boeing 777 at 103.7 dB. Private jet operators no longer appear on the airline podiums either, because flying small planes is not the same as being a quiet airline.

Measuring noise takes a microphone. Knowing whose noise it is, and whether it was noise at all, takes detective work.