The visitor ledger had been hanging for seven days when the first entry appeared.
Gyeol-ri saw it on a Tuesday morning, arriving before the tteum-jib opened to the public. She came in early most days — there was always something to adjust, some edge of a panel that wanted re-aligning, some lighting that shifted wrong overnight. The Seam district light came through the north-facing windows at an angle that changed by the week. She tracked it with a small notation in her calibration log, which she updated manually because the automated calibration system for Tier 4 galleries could not account for hand-finished panels. The system was good at measuring. It was not good at distinguishing between the kind of inaccuracy that deceived and the kind that told the truth differently.
The notebook was open.
She stood in the doorway for a moment before crossing the room. The handwriting was careful, slightly formal — the kind of script that meant someone had learned to write with attention. Not calligraphy. Just care. One sentence:
61% feels more like the kitchen I remember than any photograph I have.
No name. No date. No signature.
She read it twice without touching the paper.
The 61% panel was the one she had labeled most precisely. Burned corners in the upper left where the original photograph was overexposed. Window well reconstructed from two images taken fourteen months apart. Door frame assembled from three separate sources: one photograph from before the demolition, one from a neighbor's archive, one from the building's reconstruction filing that showed where the frame had been, not what it had looked like from inside. She had labeled the panel with a small card: 61% fidelity to source documentation. Approximation method: photographic composite, cross-referenced. She had added a second line: What the remaining 39% looked like is not known.
The Registry had flagged the panel twice during the commission review. Both flags came from the Seam Reconstruction Verification Unit — SRVU-3, which was the automated clearance system for kkaeji commissions in the Tier 4 district. SRVU-3 was not unkind. It was very precise. Its first flag read: Composite source documentation below district standard (minimum 80%). Request source supplement or fidelity reclassification. Its second flag, after she appealed with the full methodology statement, read: Appeal received. Documentation assessed. Fidelity score confirmed at 61.3%. Panel may proceed under disclosure compliance protocol. Which meant: you are allowed to make this, as long as everyone knows it is partial.
She had been fine with that. The disclosure was the work. The label was not an apology — it was the most important thing she made.
The visitor, standing in front of the panel she had fought SRVU-3 to exhibit, had written: 61% feels more like the kitchen I remember than any photograph I have.
She understood the difference between correction and response intellectually. She had been thinking about it since she hung the ledger, since she decided that blank pages were also a record, since she chose not to take the notebook down when it stayed empty for five days. The theory had seemed right. Now she had evidence.
High fidelity gives you the room. The honest approximation gives you your memory of the room.
She had suspected this. The visitor had confirmed it in one sentence, without knowing they were confirming anything.
She did not write in the ledger. That felt wrong — like answering a letter sent to someone else. Instead she went to the small desk in the back and opened her methodological notes. She wrote the sentence down with a header: Visitor response, day 7. Below that, in smaller letters: The labeled gap is not a disclosure of failure. It is an invitation to a different kind of recognition.
Then she paused and added one more line, the one she had not been able to articulate before:
The visitor did not come to see accuracy. They came to be recognized by a partial record.
She dated it. Filed it under methodology, not commission. This one belonged to the practice.
The morning got more complicated at 9:14, when SRVU-3 sent a notification to her commission terminal.
She read it standing: Compliance review initiated. 2041 Audit Protocol, Section 12(b): Tier 4 galleries with aggregate fidelity score below 75% are subject to quarterly documentation review. Your current aggregate: 72.4%. Review window: 30 days. Required materials: full source documentation for all sub-80% panels.
She sat down.
The 61% panel was the lowest, but the Mapo reconstruction brought it down. She had three other pieces in the tteum-jib with partial scores: the Yongsan corridor study (78%, almost within standard), the pre-demolition block survey from 2034 (71%, which she had disclosed in the exhibition notes but not with the same specificity as the Mapo panel), and a smaller piece from a commission two years ago, before she started the explicit labeling practice — a window reconstruction she had thought was 85% but which SRVU-3 had scored at 68% using a different methodology than hers. She had never formally reconciled the discrepancy.
The audit would find that.
She opened the compliance portal. SRVU-3's interface was clean and impersonal — the Seam district had upgraded to an integrated platform in 2039, which meant all reconstruction records lived in the same system that processed fidelity scores, issued exhibit clearances, and now apparently initiated quarterly reviews. She navigated to the methodology statement she had submitted during the Mapo panel appeal. It was still there, clear and complete. She had explained her approach: working from what the client could document, not from what was optimal. Prioritizing evidence that came from inside — personal photographs, neighbor accounts, memory-encoded visual field data where available — over external records that showed the shell of the apartment without the life inside it.
SRVU-3 had accepted this explanation for the Mapo panel.
The question was whether it would accept it for the Yongsan corridor study and the 2039 window commission, which predated her formal methodology statement and were documented under the older, less explicit standards.
She pulled up the 2039 window commission file. The client had been a family who had left the district when the first round of Tier 3 luxury developments pushed rents past what the neighborhood could hold. They had wanted the window back — not the apartment, just the window, the specific quality of afternoon light that had come through it at a certain angle in late autumn. She had worked from three photographs and a visual memory transcript from the eldest daughter, run through a commercial encoding service to produce a spatial-light estimate. The estimate was probabilistic — the service had been honest about its error range. She had used the middle estimate, which gave her a reconstruction she believed. She had not disclosed the error range in the original exhibit notes.
That had not felt like an omission at the time. It felt like one now.
She spent an hour writing a methodology supplement for the 2039 piece. She described the encoding service, its error range, the middle estimate she had chosen and why. She wrote: At the time of the original commission, disclosure standards for encoding-derived reconstructions were not yet formalized in district practice. This supplement provides the documentation that would have been required under current protocol. Then she added, because she could not not add it: The reconstruction remains an honest attempt to recover a specific quality of light that the client experienced and that is not otherwise documented. The error range does not change that intention.
She submitted the supplement to the SRVU-3 review portal. The system acknowledged receipt: Documentation received. Processing time: 5-8 business days.
Five to eight days in which SRVU-3 would work through the file and decide whether her historical reconstructions, retroactively supplemented, met the audit threshold.
She went back out to the gallery floor.
The 61% panel was getting afternoon light now. It looked different than it did in the morning — the burned corners showed less, the window well read more cleanly. The reconstruction always had a best hour. She had known this since she finished it, and she had put it in a position in the gallery where the afternoon light caught it the way morning light would never have. This was not deception. The kitchen it came from had faced southwest. Afternoon light was what it knew.
She stood in front of the panel for a while. She thought about the visitor's sentence, and she thought about SRVU-3's review, and she thought about how both were, in their different ways, assessments of the same thing: the gap between what she could document and what she had tried to make.
SRVU-3 was measuring the gap as deficit.
The visitor was measuring it as accuracy.
She understood that both were right. A 61% fidelity score was a real number. The visitor's sentence was also a real thing. They were not in conflict — they were different instruments measuring different aspects of the same object. The score told you how close she could get with the evidence she had. The sentence told you what happened when you got that close in the right direction.
The Seam's fidelity display above the gallery entrance updated once an hour. She could see it from where she stood: District aggregate: 91.4%. Her tteum-jib brought it down by a fraction she had calculated once and then stopped calculating. The display did not distinguish between the kind of approximation that deceived and the kind that told the truth differently. It was not built to.
She thought: SRVU-3 is not wrong. The display is not wrong. They are measuring something real. But they are not measuring what the visitor experienced.
She thought: the question is not whether the gap exists. The question is what you do with it.
She went back to the desk and opened a new document. Not a compliance supplement. Not methodology notes. Something else — she was not sure yet what to call it. She wrote at the top: Toward a disclosure standard for approximated reconstructions. Then below that: The current framework treats sub-standard fidelity as a compliance problem to be disclosed. What it does not have language for is the possibility that the gap, named honestly, is the work. She paused. Then: The visitor who wrote in the ledger was not consoling themselves for an imperfect record. They were reporting a finding. The approximation, disclosed, allowed them to locate their own memory in the space between what I could document and what the kitchen actually was. That location is not available in a high-fidelity reconstruction. A 95% panel gives you the room. The 61% panel gave someone their memory of the room.
She wrote for forty minutes, faster than she usually wrote, without stopping to revise. When she finally looked up, the afternoon light had shifted off the 61% panel and onto the Yongsan corridor study.
She saved the document. She would not send it anywhere today. It was a working hypothesis, not a submission. She wanted to sit with it until she knew whether it was true.
At 4:30, a second visitor arrived. Gyeol-ri was at the calibration desk when she heard the door, and she stayed there, working. She did not watch visitors — it felt like surveillance, and surveillance changed what people did with their eyes.
After fifteen minutes, she heard footsteps crossing toward the back. She looked up.
The visitor was a woman in her forties, wearing a district maintenance badge that meant she worked somewhere in the Seam infrastructure. She had been standing in front of the 61% panel. Now she was standing in front of the visitor ledger.
She was reading the entry from that morning.
Gyeol-ri watched her read it. She did not read it quickly. Then the woman put her hand in her pocket, took out a stylus, and added something below the first entry.
She did not look at Gyeol-ri. She capped the stylus, put it back in her pocket, and walked out.
Gyeol-ri waited until she was gone. Then she crossed the room.
The second entry read: I thought I was the only one who remembered it that way.
She stood there for a moment. Then she picked up the notebook carefully and brought it to the desk.
Two entries. Two strangers. The same panel.
She opened the methodology document again and added one line at the bottom: The approximation did not give one visitor their memory of the room. It gave two strangers a shared experience of having remembered something they thought they remembered alone.
She put the notebook back.
Outside, the district aggregate updated: 91.4%. The reconstruction crane on the Mapo lot had stopped for the day. The Seam was settling into evening. In thirty days, SRVU-3 would issue a review decision that might require her to reclassify her panels, update her disclosures, or in the worst case retrofit her methodology statement retroactively across every partial piece in the gallery.
She was not not worried about that.
But she was also thinking about the third person who might come, and what they would write, and how many people had been carrying a 61% memory and not knowing there was a word for it.
The notebook waited. The gallery was quiet. The afternoon had cost her something and given her something, and she did not know yet which was larger.
She turned off the calibration desk light and locked up for the evening.