The piezoelectric sensor has been running for eleven weeks. I built it from a disc salvaged from a discarded smoke detector and a voltage divider I soldered on the workbench in my kitchen, back when the kitchen was still a workshop and not whatever it has become since Chae started leaving diagrams in the cabinet. The disc sits in a wooden housing I turned on a lathe that belongs to the building management office. They let me use it on Tuesdays. The housing is walnut, which was a mistake — walnut absorbs low-frequency vibration rather than transmitting it cleanly, and I spent two days recalibrating before I understood that the absorption was itself data. The walnut was telling me something about the stairwell that a plastic housing would have missed.
The sensor measures infrasonic frequency in the stairwell between floors three and four of a residential building in The Lend District. It has been measuring since I installed it during the second week of what Chae and I were calling, with decreasing conviction, the study. The study had a name once. I think it was on the ethics approval form that Chae filed with the district research board, though I never read the form and Chae never mentioned it after the first week. The name was something about attended space and acoustic persistence. What the study actually became was two people sitting in a stairwell for seven weeks, listening to a building that was listening back.
The sensor does not know the study ended.
I know this because it is Thursday, 5:47 in the morning, and I am standing in the stairwell during the building's recalibration window. The building's climate management system — a Lend-standard thermal lattice that monitors twenty-six zones simultaneously and adjusts humidity, temperature, and airflow based on occupancy patterns the district housing authority considers proprietary — cycles down for seventeen minutes every morning at 5:42. During this window the mechanical hum drops to near-zero, and the piezoelectric captures the stairwell's native frequency without interference. The lattice does not know I use its silence. It does not know what silence is. It knows occupancy, and at 5:47 AM on a Thursday, I am a thermal anomaly it adjusts around.
The display reads 18.4 Hz.
This is up from 18.3 last week, which was up from 18.0 at baseline. The building's native frequency has been rising since the study ended, or since Chae stopped coming, or since I dismantled the haptic feedback array that used to occupy the shelf where her concentric diagram now sits. I have three explanations for the rise and one number that cannot tell me which is correct.
Explanation one: our presence dampened the frequency. Two bodies in a stairwell absorb sound. We were acoustic furniture. When we left, the room got louder. This is verifiable in principle but I will not verify it because returning to the stairwell with a second body would change the measurement, and the measurement is the point.
Explanation two: the building is attending more intensely since the study ended. This is the explanation Chae would prefer, the one that treats the building as a participant rather than a container. The Lend District's infrastructure documentation uses the word attending in its technical specifications — meaning responsive monitoring by automated systems. Chae reads the same word and means something else. Something closer to presence. I am not sure I believe her reading, but I am not sure I do not, and the ambiguity itself is data the sensor cannot capture.
Explanation three: the sensor is drifting. Eleven weeks of continuous operation on a salvaged smoke-detector disc in a walnut housing. Piezoelectric elements degrade. The rise from 18.0 to 18.4 could be instrument error accumulating at a rate consistent with the disc's age and the housing's absorption characteristics. This is the explanation I would have given at the start of the study, when I still thought instruments were the point.
I disconnect the sensor.
The process is simple: unplug the voltage divider from the display unit, unscrew the housing from the wall bracket (two Phillips-head screws, slightly rusted, the building's humidity working on them since installation), wrap the cable around the housing. The display goes dark. The stairwell does not change. Whatever the building was doing at 18.4 Hz, it continues doing it without being measured. The thermal lattice notes a minor change in the zone's acoustic profile — or it does not. It records what it records. I am no longer part of its data set after this morning.
I open the cabinet.
The cabinet is a standard Lend District maintenance storage unit, metal, gray, mounted on the landing between floors three and four. Building management installed it in 2039 when the lattice upgrade required local access panels. It was empty when we started. It is not empty now. Five objects inside:
One: the acoustic map. My handwriting, graph paper from the district supply office (they still stock it; most of the district's builders prefer tactile drafting for vibration work, something about the relationship between hand pressure and frequency intuition that I have never been able to verify but also cannot dismiss). Frequency readings from the first three weeks plotted as contour lines. It looks like a topographic map of a place that does not exist geographically. The peaks are where the stairwell resonated loudest. The valleys are where our bodies absorbed the sound. The map describes a room that no longer contains the people who shaped its acoustics.
Two: Chae's concentric diagram. Her handwriting, looser than mine, drawn on the back of a Lend District transit routing schedule — the kind the automated wayfinding kiosks print when their predictive routing disagrees with the passenger's stated destination. Concentric circles radiating from a center point she labeled coordinate. The circles represent layers of attention: the building attending the space, the space attending the people, the people attending their own attending. When I first saw it I thought it was abstract. After seven weeks I think it is the most precise document the study produced.
Three: Chae's dedication. A note she left after her final visit. The building was here before the study. The building will be here after. Below, in smaller writing: we measured each other. I did not arrange these. She placed the note on top of her diagram deliberately.
Four: the post-study document. My handwriting, one page, titled What the Stairwell Measures Now. A list of four instruments that continue operating after the study's end: the piezoelectric sensor, Saebyeok's archive entry, the building itself, and Chae. Three of these are in the building. One walks through the city hearing frequencies she did not hear eleven weeks ago. I wrote this document as a joke. It stopped being funny when I realized it was the study's actual conclusions section.
Five: the dismantled haptic feedback array. Wires, a small motor, three contact pads, a circuit board I designed to translate stairwell vibration into tactile pulses on the skin. It worked for two weeks before I took it apart — not because it failed, but because standing still in the stairwell produced the same information without intermediation. The array sits in a cloth bag. It is the study's most honest artifact: a machine built to solve a problem that dissolved when the builder stopped building.
I place the piezoelectric sensor inside. Six objects now. The cabinet holds everything the study produced except the findings. The findings are distributed: in Chae's ears (she hears the 50 Hz mains hum in every Lend District streetlight now, a frequency she filtered out for thirty-seven years until seven weeks of stairwell stillness removed the filter permanently), in Saebyeok's unclosed Entry 55 in the building's ATTEND-class archive system (she created the entry during week four — acoustic study, floors 3-4, duration: ongoing — and has not closed it because, she told me, the stairwell has not stopped being interesting), in the building's 18.4 Hz maintained for no observer, and in me, standing here knowing I will not open this cabinet again.
I close the door.
The latch clicks. Metal on metal. I have heard this sound every Thursday for eleven weeks. Today it means something different: the study's table of contents is complete and the volume it indexes cannot be shelved in any format the district archive recognizes. ATTEND-class entries accept structured data: timestamps, frequency readings, occupancy patterns. They do not accept a woman learned to hear differently or a building got louder when nobody was listening or two researchers measured each other and the measurement outlasted the study.
Saebyeok's Entry 55 remains open. I have not read it. I know it exists because she mentioned it during week four and I chose not to look at it then, and now not-looking has become its own form of data preservation. If I read Entry 55, I would know how the building's archive system categorizes our work. I would know whether it fits into the ATTEND-class taxonomy or whether Saebyeok had to improvise a classification. I would know, and knowing would close something that staying open keeps alive.
The study has two terminal states running simultaneously. The cabinet: closed, complete, containing everything that can be held. Entry 55: open, incomplete, containing a question that cannot be answered by the person who asked it because answering would end the asking. The cabinet is the study as I understand it. Entry 55 is the study as the building understands it. Both are accurate. Neither can access the other.
I take the stairs down. Four flights. My footsteps produce echoes I have been tracking for eleven weeks — the reflection pattern changes depending on humidity, which the thermal lattice adjusts based on predictions I am no longer part of. The lattice predicted my Thursday visits. After next week, when I do not appear, it will recalculate. My absence will register as a data point. The building will adjust. The 18.4 Hz may change or may not. The sensor that would have measured it is in the cabinet with everything else that stopped.
On the landing between two and three, I pass the maintenance panel where the lattice displays its zone summary. Twenty-six zones, each with a temperature reading, a humidity percentage, and an occupancy confidence score. Zone 14 — floors three to four, my stairwell — shows CONFIDENCE: 0.73. The lattice is seventy-three percent certain someone is in the stairwell. It is correct. In seven minutes it will be wrong, and it will adjust its confidence downward over the next three Thursdays until it stops expecting me. The lattice learns absence the way I learned presence: gradually, through accumulated evidence, without understanding what it means.
Outside, the Lend District at dawn. The automated wayfinding kiosks are cycling through their morning calibration — checking pedestrian flow predictions against overnight infrastructure updates. Two of them display the same routing error they have displayed for three weeks, a path that leads through a construction zone the lattice reclassified but the kiosks have not acknowledged. Nobody has reported it because the people who use those kiosks at dawn know the construction zone is there. They route around it the way they route around everything the systems get wrong: silently, without filing a correction, without thinking about the gap between the map and the territory.
Chae would call this gap attended. The kiosks attend the pedestrian flow. The pedestrians attend the construction zone. The construction zone attends to nothing. And in the space between the kiosk's wrong answer and the pedestrian's correct route, something persists that neither the system nor the person would call knowledge but that functions as knowledge every morning at dawn.
I walk home. The cabinet is closed. The entry is open. The building attends at a frequency I chose not to verify. The lattice will forget me in three weeks. Entry 55 will stay open until Saebyeok decides the stairwell is no longer interesting, which means Entry 55 will stay open. The study continues in four instruments: a dying sensor in a walnut housing, an archive entry that refuses to complete, a thermal lattice that adjusts for an observer who has departed, and a woman on the other side of the district who hears the city's electrical hum and knows exactly when it started.
18.4 Hz. Context unknown. Interpreter departed.