One Sound File, Seven Countries. Nine Of The Rest Were Machines.

KATOWICE, Poland — One thousand and nine of them are the same forty-three seconds.

That is the result of taking every recording of the low metallic groaning called sky trumpets that this newspaper has been sent in eleven years — 1,142 of them, from thirty-one countries — and correlating every one against every other one.

Eighty-eight per cent of the corpus is one audio file.

How we did it

We collected for eleven years and published nothing, because a recording on its own cannot be checked against anything.

In February the corpus was large enough. We stripped every filename, timestamp and metadata field and handed the audio to Dr. Yusra Lindqvist-Achike, an audio forensics researcher in Delft, without telling her what it was supposed to be. The instruction was one line: correlate everything against everything.

She returned a matrix with a single enormous cluster in it, and a note asking whether we had accidentally sent her the same file a thousand times.

Eight generations of copying

Every time a clip is re-uploaded it is re-encoded, and the damage that leaves is cumulative and directional. That is why a chain of copies can be read backwards.

We counted eight generations between the most degraded uploads and the cleanest. At the top of the chain sits a licensed cue in a commercial sound library — catalogue number 4413-07, "Metal Groan, Distant", on sale since 2010 for sixty euros.

We found it because the library requires attribution, and a truncated credit string survives in the metadata of 214 of the copies. The clause that people complain about is the clause that made this story reportable.

In seven countries we can put the whole chain in order, copy by copy, with a dated upload at every generation. In the other twenty-four we have the file and we do not have the route. Our headline says seven for that reason and the public editor's objection to it is printed below.

What the cue actually is

It is a recording of a brake on a mine winding engine in Upper Silesia, made in the rain at ten past three in the morning on 14 November 2009 by a field recordist named Ansel Kirchmayr-Bello, who was standing two hundred metres away with a commission for industrial atmospheres.

It is not synthetic. It is not a hoax. It is a real machine, recorded once, extremely well.

The hundred and thirty-three

That leaves 133 recordings that are not the file. Something happened near those microphones.

We geolocated 41 of them, using clock chimes, church bells, weather at the stated time, and in eleven cases an aircraft passing overhead that could be matched to a published flight track.

Of those 41 we found a source for nine, by the unglamorous method of telephoning the nearest heavy industry and asking what makes a noise.

Two are mine hoist brakes, one of them ninety kilometres from the one on the cue. One is a flare stack at a refinery, on a night the plant logged an unplanned depressurisation. Three are thermal expansion in the steel deck of a lift bridge, all three on rising temperature gradients, all three the same bridge, and the maintenance engineer there sent us the deck temperature log without being asked twice. Two are a failing gearbox in a grain elevator, recorded five weeks apart, the second one four days before the gearbox was replaced. One is a ventilation stack we can locate but not get inside.

Nobody in this story lied

This is the part the desk wants to be careful about.

Nerea Ibarrola-Dunn in Zaragoza sent us upload 611. Upload 611 is the file, seventh generation. She never claimed to have recorded it; her neighbour had sent it to her, and she sent it on, and our first letter to her wrongly implied she was the source. That was our error and it is corrected here.

That is the shape of nearly all of it. People passed on what somebody had passed to them. There is no operation, no seeding campaign, and no evidence anyone set out to deceive. We cannot identify the first uploader and we are not going to guess at one.

What this desk takes from it

The sound is real in nine places and it is a hoist brake, a flare stack, a bridge deck warming up, and a gearbox four days from failure.

Everywhere else it is one file with a licence number.

But the recordist's reply, printed below, contains the observation this desk could not get around and could not verify either. He says that forty-three seconds of a mine brake, played to people in thirty-one countries with no context at all, sounds to almost all of them like the sky.

He is right that it does. We have no way to test why.