Eleven Seconds
A sixty-second recording containing seventy-one seconds of a clock. The intervals are all one second. The calibration tone in the same file is exactly one kilohertz.
The Survey attended a routine domestic referral: sounds on the landing, nothing seen, no distress, an occupier who described herself in the first minute as almost certainly wasting our time. We measured, found a floor of −57 dBFS and nothing else, and were an hour from packing up.
Standard practice on a domestic acoustic survey is a fixed-length capture on each floor, taken as a reference rather than in hope. Sixty seconds, no operator movement, no speech. Instrument 09 takes the length you give it and runs to it; there is no stop button, because a take you can stop is a take you stop at the interesting moment, and then no two takes are the same length and nothing can be compared with anything.
This is the landing take. It is sixty seconds long.
The clock
There is a pendulum wall clock in the hall below the landing, a mass-produced Victorian drop-dial, wound weekly, running. Its escapement ticks once per second. It is audible on the recording at about 18 dB above the floor and it is the loudest thing in the file.
The take contains seventy-one ticks.
| Time | Instrument | Reading | Entry |
|---|---|---|---|
| 22 Jan 21:47 | 09 | 60.000 s | Capture commences. Length set by the instrument, not by the operator. |
| — | 09 | 2,880,000 | Samples in the file at 48 kHz mono. Verified against the file header and by count. |
| — | 09 | 71 | Escapement transients present in the take. Counted by hand three times and by threshold detector twice. |
| — | 09 | 1.000 s | Mean inter-tick interval. Range 0.997–1.003 s across all seventy intervals. |
| — | 09 | 1000.0 Hz | Reference oscillator, recorded into the same file immediately after the capture through the same chain. |
| 22 Jan 22:10 | 09 | 60.000 s | Second capture, same position, same conditions. Sixty ticks. |
| 22 Jan 22:24 | 09 | 60.000 s | Third capture. Sixty ticks. |
Interval figures are peak-to-peak of the escapement transient, measured in the review chain at the sample level. The full interval table is in the file.
The two obvious explanations, and why they are not available
The clock ran fast. A pendulum clock can run fast; a clock with a shortened pendulum or a disturbed suspension can run fast by a large margin. But a clock running fast has a shorter interval, and every one of the seventy intervals in this file measures one second to within three milliseconds. For seventy-one ticks to fit in sixty seconds the interval would have to be 0.857 s, which is 143 ms short, and which is fifty times the measured scatter. It is not close. It is not arguable.
The recording ran slow. If the capture chain had written fewer samples than the elapsed time called for — a buffer fault, a clock domain error, a dropped block silently concealed — then seventy-one seconds of room would have been laid into sixty seconds of file, the whole take would play back about eighteen per cent fast, and the ticks would sit closer together in the file than they did in the hall. Which is exactly what we see.
Except that the reference oscillator is in the same file. The operator recorded thirty seconds of a calibrated 1 kHz tone through the same microphone, the same preamplifier and the same converter, immediately after the capture. In the file it measures 1000.0 Hz. A chain compressing time by eighteen per cent would have put that tone at 1183 Hz, and it did not.
The tone was recorded after the capture rather than during it, and a fault present for the capture and absent ninety seconds later is not excluded by that alone. But such a fault would have to have been present for the whole of the capture, because the intervals are uniform from the first tick to the last: there is no seam in the file, no region where the ticks bunch and a region where they do not. A fault that runs for the entire take and stops cleanly at the end of it is not a transient. It is a different sample rate, and the file header, the sample count and the tone all say the sample rate was the sample rate.
What was done afterwards
The instrument was bench-checked the following week against a signal generator over four hours and found to be within specification. The file was submitted to two people outside the Survey — a broadcast engineer and a lecturer in audio signal processing — with the history withheld. Both were asked what the file showed. Both said, independently, that the take contained about seventy-one seconds of room. Neither offered a mechanism.
The clock was checked against the speaking clock on the night and found to be gaining about four seconds a week, which is normal for a clock of that type and which is nothing like what would be needed.
The second and third captures, taken twenty-three and thirty-seven minutes later from the same position with the same equipment, contain sixty ticks each.
What we can say
There is a file with 2,880,000 samples in it. There are seventy-one ticks in the file. The intervals between the ticks are one second. The tone recorded next to them is the right frequency.
Three of those four statements are ordinary. The Survey has not found a way to make the fourth one join them, and has stopped putting operators on it, because the last two who worked on it both ended up doing what the operator's note above describes, which is listening to a clock forty times.
Cause not established.
A sixty-second capture at 48 kHz contains seventy-one complete cycles of a one-second clock, with inter-tick intervals of 1.000 s ±0.003 s throughout and a 1 kHz reference tone recorded through the same chain in the same file measuring 1000.0 Hz. No resampling, clock or transport fault consistent with the observation has been identified.
The Thinplace Survey is invented and so is this file. The case files are horror fiction written in the format of a field report. The instruments they describe are real and are explained in full; the incidents are not.
Fiction