The patient presses a button when they hear a tone. In booth two, for four years, the button has been pressed a quarter of a second before the tone is sent, by people who cannot possibly have heard it.
I am an audiologist. Twenty-two years, the last eleven in the same department in the same building, and most of what I do is not interesting: wax, age, a lot of men who worked in factories in the seventies and are only now admitting it to their wives.
The test itself is very simple and I need you to understand how simple, because the whole of this is inside that simplicity and there is nowhere for it to hide.
You sit the patient in a sound booth. Headphones on. A button in their hand. I sit outside at the audiometer, which is the machine that makes the tones, and I send a tone into one ear at one frequency at one loudness, and if they hear it they press the button, and the machine draws a dot. Then quieter, and quieter, until they stop pressing. That threshold is the dot that matters. Then the next frequency. Twenty minutes, both ears, a chart with two lines on it.
The machine records everything. Every tone it sent, at what level, at what time. Every press, at what time. That is not for me, that is for the audit, and it has been that way since long before I arrived.
So there is a file, for every test, with two columns of timestamps in it.
We have three booths. One, two and three, off the same corridor, all fitted at the same time by the same contractor in 2009.
I started in booth two in the January of the year my mother died, which is how I can be exact about it, and the first time I noticed I assumed I had made a mistake in the setup.
It was a woman in her eighties, a straightforward age-related loss, nothing unusual. And at four kilohertz in the left, at a level she had already failed twice, the button went.
That happens. People guess. People get anxious and press on the silence because they want to do well, and there is a name for it and a way of handling it, which is that you send some silent trials on purpose and see if they press on those too. False positives. Everybody does a few.
What I did next I only did because it was a slow afternoon.
I opened the file and I looked at the two columns, and the press was not inside the tone. It was before it. The audiometer logged the tone going out at eleven minutes forty-one point nine seconds, and it logged the button at eleven minutes forty-one point six.
Three tenths of a second early.
You cannot press a button before a sound in a way that gets recorded as a response to that sound. The machine pairs a press with the tone it falls inside. This one fell outside and it got paired anyway, because it was close enough that the software rounded it in.
I want to be honest about what I did with that, which is nothing, for about seven months.
I told myself it was a clock. Two clocks, really: the one on the tone generator and the one on the response logger, and if they drift apart by three tenths of a second then everything I just described is a maintenance job and not a story.
So eventually I did the obvious thing. I put the same patient through booth two and then booth one, back to back, on the same afternoon.
Booth one: every press inside its tone, the way it should be, for the whole test.
Booth two: four presses early. Same patient, same headphones, same me, forty minutes apart.
I had the machine in two swapped out. New audiometer, new headphones, new response button, new cable, and the estates lads pulled the floor up to check the run, and I paid attention to the dates because I wanted to know whether it stopped.
It did not stop.
Since then I have kept my own record, which I know how that sounds.
Four years. Two thousand one hundred and something tests in booth two. Early presses in about one test in nine, which is far too many to be nothing and far too few to be a fault.
Here is everything I have been able to establish, and I have tried very hard, and none of it helps.
It is always early, never late. In four years I have not once seen a press logged after its tone by more than the ordinary human reaction time. If it were clock drift it would go both ways eventually. It does not.
It is always between two and four tenths of a second. Never a second. Never half a second. A narrow window, the same window, for four years.
It happens at the quiet end. It never happens at a level the patient can comfortably hear. It happens at or below the threshold I am about to write down as the point where they stop hearing.
And it happens to people who are deaf in that ear. That is the part I cannot put anywhere. I have had it on an ear with no measurable hearing at all, a dead ear, an ear that is not doing anything, and the button went, early, at four kilohertz, on a tone that was sent into nothing.
There is one more thing and then I will stop.
I started, about two years ago, sending tones I did not intend to send.
Not really. What I mean is that the audiometer has a rehearsal mode where you can queue a sequence and step through it, and I got into the habit of queueing the next tone and then waiting, sometimes for ten or fifteen seconds, before I actually sent it. Just holding it there. There is no clinical reason. I wanted to see something.
And what I see, perhaps a dozen times now, is the button going while the tone is queued and not sent.
Not three tenths early. Nine seconds early. Eleven seconds early. Once, twenty-two.
The machine does not log those, because there is no tone for it to pair them with. They go in nowhere. They exist as a click in my headphones, which are wired to the same button so I can hear the patient pressing, and as a woman of eighty-three in a soundproof box with her hand up, saying, sorry, was that one, I think that was one.
I have stopped doing that. I do not queue any more. I send the tone the moment I decide on it and I try not to decide until my hand is already moving, and I know exactly how that sounds and I do it anyway.
Booth two is still in service. It passes its calibration every year with no comment.
I have written it up twice. The first time it came back with a note asking me to quantify the clinical impact, and the honest answer is that there is none: an early press at a level below threshold does not change the dot I write down, because I do not count a response I cannot reproduce. The chart comes out the same. The patient gets the right hearing aid. Nobody is harmed.
The second time I did not send it.
I have a colleague who has read all of it and who thinks it is a software fault in a discontinued audiometer that got carried over into the replacement by whoever wrote the driver, and I would very much like her to be right and she may well be.
But I sat in booth two myself, on a Friday evening, with the door shut and the button in my own hand, and I had a technician outside send a sequence he chose and did not tell me.
I pressed nine times. He sent seven tones.
Two of the nine went in nowhere. I do not remember pressing them. My hand pressed them and I felt it, the way you feel your own knee jump when a doctor taps it.
And of the seven I did hear, I have the file, and four of my own presses are logged before the tone that caused them.