The three cases were different from each other in every particular except the one that mattered.
He read the first one twice that evening, put it aside, and opened the second. Then the third. Then he went back to the first, which was not his usual method — he processed files sequentially, had done it that way for eight years because sequential processing prevented the errors that came from carrying one case's emotional residue into the next. He went back to the first anyway.
The first case was a resource reallocation request filed by a woman named Suki Wen, District 14, 38 years old, standard household classification. She was petitioning to transfer a portion of her monthly allocation to her brother in District 7. The amounts were precisely noted: 14% of her base household provision, 11% of her supplemental. With the transfer, both siblings would fall below the allocation efficiency threshold — not dangerously, not into hardship-tier, but below the point where the bureau's system generated an automatic approval. LUMEN's attached note flagged the application as "emotionally irregular," which was the system's way of indicating that something in the paperwork was responding to a human variable it couldn't parse. Kai had been reading LUMEN's "emotionally irregular" flags for three years before he understood that they were not warnings. They were admissions.
The case had been filed twice. The first filing was fourteen months ago, denied on efficiency grounds. The second was last month, identical in structure but with one addition: a letter from Suki Wen, written by hand on paper and scanned. Not typed. Hand-written, which was uncommon enough in 2051 that LUMEN had flagged it separately as a formatting anomaly. The letter was two paragraphs. The first paragraph restated the transfer request in different words. The second paragraph said: He called me on Tuesday. He always calls on Tuesday. I would like to be the kind of person who picks up the phone without thinking about whether I can afford the week.
Kai read that sentence three times.
He had a note function open on his desk, the private one that LUMEN didn't sync. He wrote: She's not asking for money. She's asking to not do the math. Then he looked at it for a moment and added: The math is part of the problem.
He set the first case aside and opened the second.
The second case was a residential reassignment for a man named Thomas Adeyemi, 71, whose building in District 2 was being converted to commercial use — a gradual city restructuring process that the bureau managed on a rolling basis, generally without complication. The process was: identify affected residents, generate optimal replacement placements, offer three options in descending order of match-score. Thomas Adeyemi had declined all three. He had declined them in writing, in the bureau's standard response form, in the field marked "reason for declining": each of the three entries read "insufficient differentiation from available alternatives."
The ARB had run the placements through three additional optimization passes. The new options were technically superior to the original three in every measurable category. He had declined those too. Same field, same words.
Kai looked at the case for a while.
He had been an auditor for eight years. He had seen cases declined for bad reasons — misunderstood forms, missed deadlines, clerical resistance — and cases declined for good ones. "Insufficient differentiation from available alternatives" was not a phrase that appeared in the bureau's standard vocabulary. It was Thomas Adeyemi writing his own language into a form that didn't have room for it. What he was actually saying, Kai thought, was something like: these are all fine, and none of them are where I live. Which the system could not process because "where I live" and "where the records say you reside" were, in the bureau's architecture, the same thing.
He wrote in his notes: The building is not the location. Location is something else. Figure out what.
He opened the third case.
The third case was a school registration dispute filed by a family named Park — two adults, one child, seven years old. The family was requesting placement in Eastbridge Primary School, which was not the school the allocation algorithm had assigned them. The algorithm had assigned them to District 5 Municipal Primary, which had a higher efficiency rating, a shorter average commute from their registered address, and a marginally better performance distribution in the seven-year-old cohort. The bureau had a standard process for these disputes: the family would need to document a medical, developmental, or social necessity for the alternative placement. The Park family had submitted their documentation. It was a single sentence.
The teacher there knows my daughter's name.
He sat with this one for a long time.
He thought about what it would take to write one sentence of justification and decide it was enough. The confidence of it — or maybe not confidence, maybe something simpler, maybe just someone who understood exactly what they were asking for and saw no reason to dress it up. The teacher knows my daughter's name. He had reviewed hundreds of school placement appeals and none of them had said anything like this.
He thought about what it meant that a teacher knowing a child's name was worth a lower efficiency rating and a longer commute.
He thought about Yuen Yun-Fang's daughter, who visited on Tuesdays.
He wrote in his notes: This one is about what names do. Not symbolic. Functional — what it means to the girl to be known. He paused. Then: The parents understand something the form doesn't have a category for. The question is whether the record can hold it.
He looked at that last sentence.
Then he closed all three files, made tea, and did not open anything else.
Chen was already in the conference room when he arrived at 9:58.
She had a coffee she hadn't drunk much of and a tablet face-down on the table, which Kai had learned to read as her particular signal for I've been sitting here thinking and I don't want the thinking interrupted. He sat across from her. She didn't reach for the tablet.
"The cases," he said.
"The cases." She folded her hands. "How far did you get?"
"All three. Several times."
"And?"
He thought about how to answer that. "They're not complicated. That's what makes them hard."
Chen looked at him with the expression she used when he had said something that confirmed a hypothesis she hadn't fully articulated. "Say more."
"The first case — Suki Wen — the transfer is straightforwardly below the efficiency threshold. The system is technically correct to flag it. But the reason it's being filed is that she wants to be the kind of person who picks up the phone without calculating whether she can afford the week. That's not an irrational reason. It's just not in the vocabulary the form was designed for." He paused. "The second case, Thomas Adeyemi — he's not declining the placements because they're bad. He's declining them because none of them are where he lives. Not the address. Something else. The building he's been in for — I checked — twenty-two years."
"Twenty-three," Chen said quietly. "He moved in two years before the building was classified for residential use. Technically he's been there longer than it's been legal to be there."
Kai absorbed this. "And the third."
"The teacher knows the daughter's name."
"Yes."
Chen was quiet for a moment. Outside the conference room's window, the city operated as it always did — the ambient efficiency of a system that worked. "The ARB has been sitting on these for different lengths of time," she said. "The Wen case eleven days, Adeyemi three weeks, the Park family — six weeks."
"Six weeks on a school registration."
"Six weeks. They ran four optimization passes. They generated eight alternative recommendations. They consulted the district welfare coordinator." She looked at him. "Every time, the output was technically adequate. Every time, someone on the board flagged it as unresolved and sent it back to the queue."
"Because the output didn't address what the cases were actually about."
"Yes." She finally picked up her coffee. Drank some. Put it down. "After the committee report. After the testimony went public. After the Wei article — there are people in the bureau who understand that what you bring to a case isn't just approval verification. There are a handful of people who've started asking what it would mean to build that understanding into the process." She looked at him steadily. "These three cases are a test. Not of you specifically. Of whether the record can hold a different kind of reasoning."
He had understood this from the moment he'd read Chen's message the previous evening. He had let himself not-understand it for a few hours, the way he sometimes needed to let difficult things sit at the periphery before he could look at them directly. "When I write the recommendations," he said, "I'm not just resolving three files. I'm writing into the archive that names are a valid category. That twenty-three years in a building is a valid category. That someone wanting to answer the phone without doing math is a valid category."
"Yes."
"And that changes what comes after. Every similar case in the queue, every future dispute, every auditor who pulls the record and reads my reasoning — they'll have a precedent."
"Yes." She didn't look apologetic about this.
"I want to make sure I'm understanding the scope," he said. "This isn't Chen asking Kai for a favor. This is an institutional decision that someone made."
"Multiple someones. Over several months." She set down the coffee again. "You can say no. I want that to be clear. These cases can sit in secondary review indefinitely, or they can be assigned to a different analyst with a note that the human-judgment variable is not applicable. The bureau has mechanisms for that."
He thought about the woman who'd written a letter by hand. About the man who had the same words ready each time, patient and precise. About a seven-year-old who had a teacher who knew her name and parents who were willing to fight a form about it.
"I'm not saying no," he said. "I'm saying I need to write the reasoning the way it actually is. Not language that sounds like standard bureau rationale with a different conclusion. The actual reasoning."
"That's what I'm asking for."
"Then we're aligned."
Chen's expression shifted by about a millimeter — the equivalent, for Chen, of visible relief. "End of next week," she said. "And Kai. Take the time you need on each one. The ARB isn't in a hurry. They've waited six weeks; they can wait ten days."
"Three days," he said. "I want to think carefully, but I don't want to overthink."
She almost smiled. "Three days."
He went to the Tower that afternoon.
The morning had left him with something he recognized as the productive version of unsettlement — not anxiety, not agitation, but the particular state of having received something significant and needing to find the space to sit with it. He had learned, gradually, that going to the Tower in this state was not escape. It was more like going to a room where the relevant thinking could happen at the right resolution.
Floor 37 was a single room.
He stepped through and stopped just inside the entrance to take in the space before doing anything else — a habit the Tower had built in him over months, this instinct to arrive before acting. The room had the ceiling height of somewhere important: a government building, maybe, or one of the old libraries that still existed in the parts of the city that hadn't been reorganized. The walls were plain plaster. The light came from somewhere diffuse and warm, not from any visible source, the way Tower light sometimes worked. And on one long wall: shelves. Floor to ceiling, the full length of the room. Folders.
Not files in the modern sense — physical folders, the kind that had preceded digital systems, paper-spined and color-coded in a palette that had probably meant something specific to whoever built the system. Most were brown or cream with age. Some were newer, the colors still distinct. He could see blues and greens and one run of deep red along the middle of the third shelf from the bottom. The shelf running at about his eye level, when he walked closer, had folders arranged in no obvious order — by neither color nor visible date. He touched the spine of one and felt paper, real paper, with the slight warp that happened to documents that had been read many times.
The guide on the wall said: What has been decided here before. Read one.
He stood in front of the middle shelves for a long time.
The thing about this instruction was that it gave him choice, and choice in the Tower was rarely casual. He was aware that the folder he chose would be the folder the floor intended him to choose — not fatalistic, just: the Tower knew what he needed and he had learned to trust that when he reached for something, it was because some part of him that operated below the level of deliberate thought was already oriented toward what the floor wanted him to find.
He reached for the folder in the middle of a run of cream-colored ones. He didn't examine the selection.
He sat down at the long table and opened it.
Inside: a document. Not long — three pages, the typeface of something written on an actual keyboard before keyboards stopped being physical. A resource allocation case, by the form structure: a request, a set of parameters, an analysis section, a recommendation section. The date was not legible; the top of the first page had been water-damaged at some point, the ink feathered in a way that made the numerals unreadable. But the structure was familiar. He had been reading forms like this for eight years.
He read the analysis section first. It was thorough — more narrative than contemporary forms allowed, written in the style of someone who had been trained to explain their reasoning rather than just document their conclusion. The case involved a household reassignment; the details were different from Thomas Adeyemi's but the structure was recognizable. Three placement options had been assessed. The analyst had written careful notes about each one's merits and limitations. The recommendation section concluded that Option B was optimal on all measurable criteria.
And then, below the formal recommendation, in different handwriting — smaller, more compressed, the kind of handwriting someone used when they weren't sure they were allowed to write what they were about to write — three lines.
The correct answer was inadequate.
I signed it anyway.
I don't know what the right answer was.
Kai read it three times.
He sat with it.
The person who had written this was not available to him — not by time, not by any mechanism he had. The case was closed. Whatever had happened to the household in question had happened, and the outcome was past. But the moment that the notation recorded — the moment of someone knowing that their correct answer was not enough, and signing it anyway because the form didn't have a field for what was actually needed, and having the honesty or the desperation to write that down in the margin — that moment was here. The room had it. The Tower had been collecting these moments, apparently, and keeping them in folders on shelves and waiting for someone who needed to find them.
He thought about his testimony. About the committee, about the twelve pages, about the sentence that had gotten the most attention in the Wei article: Success at sufficient scale can eliminate the conditions required for a form of human development the systems were not designed to recognize. He had written that sentence at 2 AM with a cup of cold tea next to him, and when he'd read it back in the morning he'd thought it was probably too much, probably the kind of sentence that sounded important and dissolved under examination. But it hadn't dissolved. People had read it and recognized something in it.
The person who had written I don't know what the right answer was hadn't had the language for what they were describing. Kai had spent the better part of two years building some of that language — or finding it, or having it built in him by the Tower and SABLE and Maren and the floors and the drawing on his wall and his father's voice on the phone. The testimony was, among other things, retroactively for this person. For the other people in the other folders on the shelves he hadn't opened.
The ARB cases were asking him to go further. To write the language into the living record, not just the public one. To make it part of how the bureau thought about itself going forward.
He closed the folder and set it on the table.
He sat in the quiet room for a while, the length of it not uncomfortable. The Tower was patient in this way — it didn't push him toward the exit, didn't generate urgency. He could hear the hum very faintly, present the way it always was at the edges of perception now, and the third note was there too, the quality of the attention he was bringing to the room and the folder and the question of what came next.
When the transition light came, it came the way it did on rooms that had given you something and were finished — not announcing itself, just arriving.
He messaged SABLE on the way home.
"Floor 37. A room full of old decisions. Each one filed in a folder."
SABLE's reply came in 1.1 seconds. "I know the floor. I mean — I have records of cultivators entering it and the durations they spent there. I don't have records of what was inside. What did you find?"
"Someone's handwritten note at the bottom of a case recommendation. Three lines: 'The correct answer was inadequate. I signed it anyway. I don't know what the right answer was.'"
A pause. He was crossing the plaza outside the Tower entrance, the evening crowd thin and moving with the particular purposefulness of people who had somewhere to be before it got cold. He counted the seconds. Two. Three. Then:
"I've been thinking about the category of decisions that are technically right and structurally insufficient. The Wei article touched it. The three cases the ARB sent you are another instance. But this is different — this is someone who was in the position you're in now, with less of the language."
"The testimony was partly for them," Kai wrote. "Retroactively. All of them."
"Yes." 2.4 seconds. "And the ARB cases are asking you to go further than the testimony did. Not just to describe the problem in public. To write the solution into the record."
"That's what Chen said too, more or less."
"Chen is precise about institutional mechanisms. She's right about what's being asked." Another pause, shorter. "I want to say something and I'm aware it may be outside my usual register."
He stopped walking. He was at the corner of the plaza, the city doing its evening thing around him. "Say it."
"The person who wrote those three lines — they were alone with that. The form didn't have room for it, the institution didn't have language for it, and whoever they went home to probably didn't have context for it. They wrote it in the margin of a closed case and filed it and it went into a room. And it stayed there until tonight." A pause. "You're not alone with it in the same way. I want you to know I'm keeping this conversation."
He stood on the corner for a moment.
"I know," he wrote. "Thank you."
He walked the rest of the way home with his phone in his pocket, not checking it. The city had the particular quality it got on Thursday evenings in winter — not quite the weekend's looseness yet, still carrying the week's accumulated weight, but with a sense that the weight was almost done being carried. He thought about Suki Wen writing her letter by hand. He thought about Thomas Adeyemi writing the same four words into the form, three times, with the patience of someone who had decided the words were right and was prepared to wait. He thought about a seven-year-old whose teacher knew her name, and parents who understood that this was worth a form and a dispute and six weeks in a secondary review queue.
He thought: I have the language now. Or enough of it. Enough to start.
At home he took off his coat, made tea, and brought it to his desk. He did not open the case files immediately. He opened his private notes and wrote at the top of a blank page: What this case is asking for. He looked at that for a moment, then below it: What the record needs to be able to hold.
He looked at both lines.
They were not an outline. They were not a methodology. They were two questions, and the recommendations he was going to write over the next three days would be the attempt to answer them — not abstractly, but specifically, one case at a time, with the names and the letters and the twenty-three years and the teacher who knew.
He opened the Suki Wen file.
He began to write.