Seven levels, seven sets of landing doors, seven positions the controller knows by a magnet on the guide rail. The event log has a position between six and seven, and the car has stopped there nineteen times.
I am a lift engineer. Twenty two years, mostly traction lifts in hotels and offices, and the one in this story is on my monthly list.
You need to know how a lift knows where it is, because it is the whole of this.
It does not know by time or by motor turns. On this machine, which is a fairly modern traction lift, there is a steel tape and a set of magnets on the guide rail in the shaft, one at each floor level, and a reader on top of the car. The car passes a magnet, the controller counts it, and the controller knows which floor it is at. There are also separate slowdown and final limit switches, and a separate absolute position encoder on the governor, and the two have to agree or the lift takes itself out of service.
That redundancy is not belt and braces for comfort. It is because the landing doors can only unlock when the car is in the door zone, and the door zone is defined by that position. A lift that is wrong about where it is opens a set of doors onto a shaft.
So the position is the most carefully guarded number in the machine.
The building has a basement, a ground floor and floors one to five. Seven levels, seven sets of landing doors, seven magnets.
The controller logs every journey: time, from, to, direction, load, door cycles, and any fault. I download it every month onto a laptop because reading the log is most of a service.
Nineteen times since 2019, the log has a journey ending at a position the controller records as between six and seven.
I should explain what those numbers are. Internally the floors are indexed one to seven, basement being one. So six is the fourth floor and seven is the fifth.
The log does not record fractional positions. It cannot. The field is an integer floor index, and every other line in eleven years of that log is an integer.
Nineteen lines have a value in a different field instead: the raw encoder count, with no floor index at all, and the raw count sits between the count for six and the count for seven. Closer to seven. About four fifths of the way up.
The controller logged those journeys as completed. Doors opened. Door cycle recorded, both leaves, open and close, with the times you would expect.
The shaft between the fourth and fifth floors.
I have been in it more times than I can count and I have been on top of the car through that section with a torch at service speed, which is the slowest and most boring way to see a shaft and the only way to see it properly.
There is nothing there. Brick on three sides, the guide rails, the counterweight running in its own guides, the travelling cable, a junction box at fifth floor level and the fourth floor landing door at the bottom of the section.
No fifth set of doors. No opening. No bricked up opening, which is the obvious thought in an old building, and this is not an old building, it is 1987 concrete frame with a blockwork shaft, and the blockwork is continuous and I have had a lamp along every course of it.
What gets ruled out by the door cycle, which is the part that made me stop treating this as a data fault.
A corrupted log entry is a real thing. A position that reads wrong because the reader is dirty or a magnet has shifted is a real thing and it is most of my job.
But a wrong position on its own does not open a door. The door operator is a motor on top of the car and it has its own encoder, and it logs a door cycle when the doors physically move, with the current draw and the time, and you can see from the current trace whether they met an obstruction.
All nineteen have a full door cycle. Open, dwell between eleven and forty seconds, close. Normal current. No obstruction, no reversal.
And the landing doors cannot open unless the car is at a landing, because the car has a mechanical vane that engages the landing door lock, and the vane engages a roller on the landing door, and those rollers only exist at the seven landings.
So for nineteen of those cycles, the car doors and a set of landing doors both opened, four fifths of the way between the fourth and the fifth floor, where there are no landing doors and no roller for the vane to engage.
What was done, by the company, properly.
The tape and the magnets were replaced in 2020. The reader was replaced. The position encoder on the governor was replaced in 2021 and the controller board in 2022, under a goodwill arrangement because by then I had reported it four times and the manufacturer's technical man had been out twice.
His report is two pages and I have a copy. The conclusion is that the events are not reproducible, that the safety circuit has been tested and found compliant, and that the lift is safe to remain in service.
Every sentence in that is correct, and the lift is safe. I would ride it. I do ride it.
The times.
All nineteen are between 02:00 and 05:00.
Seventeen of them have a load recorded. The controller weighs the car, because it has to, for the door holding and for the overload alarm.
The loads are between sixty and eighty kilos. One person.
Two of the nineteen show no load at all, and those two are the ones I would leave out if I were trying to make this sound better than it is, because an empty car doing something odd is easier to live with.
The one thing I have done that is mine and not the company's.
In 2023 I put a camera in the car. Not a hotel camera. My own little battery one, up in the corner by the ceiling hatch, pointed down, which I am not supposed to do and which I took out again after six weeks.
In six weeks it caught one.
The log says 03:41, position between six and seven, load seventy two kilos, door cycle nineteen seconds.
On the video the car is empty. It is empty when the doors open and it is empty for the nineteen seconds and it is empty when they close, and past the open doors you can see what is out there, and what is out there is lit.
Not shaft. Not brick. A corridor, carpeted, with a wall light on the right about head height and a door frame further along, and it is the wrong carpet. Ours is a dark blue with a pattern. That is a plain carpet, pale, and the light is a round glass shade on a short brass arm, and there is not one fitting like that anywhere in this hotel.
I did not put the video in any report and I took the camera out the same week.
What I did instead is a thing the log will show forever, which is that on that service visit I adjusted the levelling on floors four and five, which needed doing by about three millimetres and which I noted honestly.
And I moved one magnet on the guide rail, by about forty centimetres, up, so that the stretch of shaft where this happens is now inside the door zone the controller believes belongs to the fifth floor.
It means that if it happens again, the controller will record it as an ordinary stop at the fifth floor, and no engineer reading that log will ever see anything at all.
I have turned that over for two years. I am not sure it was the right thing to do and I am completely sure I would do it again, because the alternative was to keep filing reports until somebody senior enough came out at three in the morning and rode it down with the doors open to have a look.