The first morning of writing, he started with Suki Wen.
He had the file open, her letter scanned and centered on his screen, the handwriting visible in the way handwriting is visible when you haven't seen much of it — the particular pressure variations, the places where the pen had lifted, the word Tuesday that she had written and then rewritten over so slightly, as if she'd started with one version and changed it before she finished. He had enlarged the scan to see the detail and then reduced it again, feeling obscurely that reading someone's handwriting too closely was a different kind of attention than the situation required.
Her argument, in two paragraphs, was simple. She wanted to transfer part of her allocation to her brother. The amounts would take both of them below the efficiency threshold. She had filed this once before, been denied, and filed it again with the letter. The letter didn't argue against the denial; it didn't cite regulations or precedents or exceptional circumstances. It said: He always calls on Tuesdays. I would like to be the kind of person who picks up the phone without thinking about whether I can afford the week.
Kai had a blank document open for the recommendation. He looked at it for a while without writing anything.
The bureau's standard recommendation structure had five sections: Case Summary, Relevant Parameters, Variance Analysis, Assessment, and Determination. He had written hundreds of them. He could write the first three sections in his sleep — the summary was the facts, the parameters were the thresholds, the variance analysis was where Suki Wen's request diverged from them. All three sections would be accurate and would not be the point. The Assessment section was where he had to find the language.
He started typing.
The requesting party is not in dispute with the efficiency threshold as a principle. She is not arguing that the threshold should not exist or that her household's allocation should be recalculated. She is asking the system to approve a specific action that falls below the threshold on grounds that the action matters to her in a way the threshold does not measure.
He stopped. Read it back. Deleted the last clause and tried again.
She is asking the system to recognize that the benefit of the transfer is not captured by the same metrics as its cost. The cost is economic — two households below the threshold. The benefit is not economic, which means the allocation system has no mechanism for valuing it. This is not a design error. It is a design boundary.
He sat with design boundary for a moment. It was the right phrase. More neutral than failure, more honest than limitation — it named the thing without apologizing for it.
When a petitioner asks for something whose value the system cannot measure, the standard process is denial: if the value cannot be demonstrated within the system's metrics, the system treats it as zero. But value that the system cannot measure is not the same as value that does not exist. This case asks whether the record can hold that distinction.
He stopped again. That was the sentence. He read it several times. Value that the system cannot measure is not the same as value that does not exist. He had been thinking this for months — had thought it in relation to Case 7743, had thought it on Floor 7 with the drawing, had thought it in the rain outside the Tower after Floor 21 — and this was the first time he had written it in a document that would enter the bureau's formal archive.
He continued writing.
The Suki Wen recommendation took him most of the morning. Not because it was long — the final document was four pages — but because every paragraph required him to find language that was accurate to the bureau's register while being honest about what was actually being claimed. The bureau's register tended toward clinical nouns and passive constructions: requests have been received, assessments have been conducted, determinations have been made. He kept this syntax and changed what it was saying. The effect was slightly odd, a familiar form holding unfamiliar content. He thought Chen would read it and know exactly what he had done.
The determination at the end was: Approve. The transfer is below the efficiency threshold by a margin insufficient to constitute hardship risk. The benefit to the petitioner is not economic but is real and specific. The record should hold that the system approved this request on the basis of a value it cannot measure, and that this approval was made in full knowledge of that fact.
He sent it to his private backup, made tea, and opened the Adeyemi file.
Thomas Adeyemi had lived at 14 Meridian Court, District 2, for twenty-three years.
Kai had read this in the file. He had also called the district records office that morning before the writing session, a thing he did not normally do — his cases generally came with complete documentation — but the Adeyemi file had a gap he wanted to close before writing. The building had been in a transitional classification since the redevelopment order; the records office had a note in the file that the order had originally been appealed by a tenant committee and the appeal had taken fourteen months before being withdrawn. Thomas Adeyemi was listed as one of the committee's three organizers.
He had appealed. He had spent fourteen months in the formal dispute process. And then he had withdrawn the appeal and begun declining the replacement options instead, which was a different kind of response to the same situation — not the institutional response, the personal one. Insufficient differentiation from available alternatives. He had written those words four times now. Kai had read them four times and each time thought: he already knew the appeal wouldn't change anything. He just wanted to see if the system would listen. It didn't. So he started saying, in the only language the form had given him, what he actually meant: these alternatives are insufficient because they are not where I live.
He took the afternoon and walked to Meridian Court.
This was not standard practice. He had never visited a case address in eight years of auditing. The bureau did not require site visits; the bureau did not, in its architecture, require anything that looked like it might produce a non-data relationship between the auditor and the case subject. He went anyway. Not to meet Thomas Adeyemi — he had no authorization for that, and the ethics of it were uncertain — but to see the building.
It was a mid-century structure, eight floors, brick facade darkened with forty years of city air. The proportions were of its era — slightly taller ceilings than modern buildings, windows that were a few centimeters wider, the kind of dimensions that predated the optimization of human habitation. There was a courtyard visible through the building's ground-level archway, planted with something that had been tended — not formally, not according to any landscape design he recognized, but consistently, the way things get tended when someone has cared about them for a long time. He stood outside for perhaps ten minutes.
He looked at the building and thought: the replacement options were in District 2, same neighborhood, modern, well-rated. They were probably better buildings by every metric the bureau used. They were not this building. This building had a courtyard with plants that someone had tended for twenty-three years. The records would not show that. The site visit would not show it either, except that he had now seen it and it was in him as information in a way it hadn't been when it was in the file.
He walked back to his apartment and wrote the Adeyemi recommendation.
He thought, writing it, about what it meant that the site visit had changed his understanding of the case. Nothing in the file had changed. The data was the same. What had changed was that he had stood outside the building and let it register — not the information, the thing the information pointed toward. The courtyard plants. The proportions of the windows. The way forty years of city air had darkened the brick in a particular pattern that was not the same as any other building's pattern. He had let this matter to him, and then he had written a recommendation that was better for it. He thought: this is what the second reading is for. Not to find errors. To let the person become real enough that the recommendation is actually about them.
The language he found: A resident of twenty-three years has developed a relationship with a location that is not captured in registered address records. This relationship constitutes a form of investment — in time, in community, in the accumulated texture of a particular life lived in a particular place — that the allocation system's replacement process does not currently account for. The three provided alternatives are objectively adequate. They are not, by the petitioner's assessment, adequate to what he is being asked to leave. This gap is not irrational, and treating it as irrational is not a service the bureau should provide.
Determination: Approve the petitioner's request for a delay of six months in the replacement timeline, during which the bureau will work with the petitioner to identify a replacement that addresses the non-economic value at stake. If no such replacement can be identified, the case returns to standard process.
He read it back three times. It was not the simple approval he had started thinking it would be — it was a different kind of answer, one that acknowledged the gap without pretending to close it. He thought it was honest.
He went to see Maren on the second evening.
She had texted him that afternoon — not with a reason, just: Are you around later? — and he had replied yes, and they had agreed to meet at the plaza outside the Tower, which was where they tended to meet when neither of them was going inside. He arrived at seven and she was already there, sitting on the low wall near the entrance with a coffee she'd brought from somewhere, watching the people going in.
He sat next to her and they watched the Tower entrance for a while without talking. This was a thing they could do — had been able to do for some time, since well before he had understood what their particular friendship was made of. The silence between them was not the kind that needed to be filled.
"You're writing those ARB cases," she said finally. Not a question — she had heard about them through whatever network Maren heard things through, probably Juno or someone else in the upper floors community.
"Second day. Two done, one left."
"How is it going?"
"Better than I expected. The language is there when I look for it. I wasn't sure it would be."
She nodded. She drank her coffee.
He was aware — had been aware, off and on, for months — that there was something between them that had never been named and that the not-naming of it had served some purpose he hadn't fully understood. It was not the same kind of thing as what he had with Lin. What he had with Lin was forward-facing, relational, alive in the present and oriented toward the future. What he had with Maren was different in structure — it was more like what you had with someone who had been exactly where you were before you got there, who had walked the terrain in advance and said nothing about it except you'll find it when you get there, which had been more useful than any map. He had thought sometimes that this was the entirety of it. That the unnamed thing was simply: she had gone first, and he had followed, and the bond was the path.
"I wanted to tell you something," she said. "I've been thinking about how to say it for a while."
He waited.
"When I first talked to you — Floor 8, the village — I was ahead of you, and I knew it, and I was careful about how I used that." She turned to look at him directly, which she did rarely, the particular quality of her directness that had always registered as honesty rather than confrontation. "Being careful was the right call. You didn't need me to walk you through it. But I was also —" she paused, not looking for the word but making sure of it before saying it. "I was also watching to see if you'd catch up. And that was about me, not about you."
"Catch up to what?"
"To where I was. Which isn't — I don't mean that the way it sounds. It's not about the floors. It's that I thought you were a version of a thing I had already seen, and I was curious to see how far you'd go, and that's not the same as just seeing you." She drank the last of her coffee. "You went further than the thing I'd seen before. And at some point I stopped watching to see if you'd catch up and started just — watching. Because you were doing something I hadn't seen done that way."
He thought about this. The Tower entrance was busy at this hour — early-evening climbers, people finishing their working days and arriving with the purposeful quiet of regulars.
"The testimony," he said.
"Partly. But before that. The way you moved through the Tier 1 floors — there was something in it that was yours, not the pattern I'd been comparing you to." She looked at the Tower. "I think what I'm saying is: I've been treating our friendship as a kind of observation study for longer than I should have. And I wanted to stop doing that. So I'm telling you."
He sat with this for a moment.
"I didn't feel observed," he said. "That's not — I'm not saying it to be kind. I actually didn't. It felt like something else."
"What did it feel like?"
He thought about the first time she had appeared at Floor 8 and sat next to him and said you're doing it right. About Floor 12 and her waiting outside when he emerged. About the conversation on Floor 17 and the way she had listened without offering anything. About Maren's Tier 2 cost — the certainty she'd thought was wisdom — and the message she'd sent him the same afternoon.
"It felt like having someone who understood the territory," he said. "Not a guide. More like — someone who had been through the same country and left the landscape more readable by having been there."
She was quiet.
"That's generous," she said.
"It's accurate."
They sat for a while longer. The Tower entrance cycled through a small rush and then quieted.
"I have Lin now," he said — not as a non-sequitur, though it might have sounded like one, but as something that belonged to the conversation. "That's — named, in a way things between us weren't. And I wanted you to know that what we are to each other is not less to me because of that. It's just different in kind."
Maren was quiet for long enough that he wondered if he had misjudged — if the naming had been too explicit, if she had meant something he hadn't read correctly. Then she said: "I know. And what I wanted to tell you goes both directions." She looked at him. "I wasn't only observing. I was also — there were things that happened in my climbing that I didn't tell anyone else. That I told you. Not as a counselor or as a subject. As —" she stopped. Tried again. "As someone I trusted with the real version."
He understood this. He thought he had known it, in the way you know things that are true before you have the words — the same way the Tower floors worked, the knowledge preceding the naming.
"That doesn't have a category," he said.
"No," Maren agreed. "It doesn't."
"Maybe it doesn't need one."
She looked at him with the particular expression she used when he had said the right thing in a way she hadn't expected. "Maybe it doesn't." A pause. "How's Floor 37?"
"Cleared. A room of old decisions — someone's handwritten margin note. It stayed with me."
"That floor is hard to let go of." She stood, tucked the coffee cup under her arm. "I'm at 43 now. The floors up here are — different. The Tower stops testing whether you can hold things and starts testing what you'll do with what you're holding." She looked at the entrance. "Good luck with the Park case."
"The teacher who knows the daughter's name."
"That one will write itself," she said. "Once you find the first sentence."
She left. He sat on the wall for a while longer, the Tower's hum present and low, the third note in the quality of the air around the building at this hour — the accumulated attention of everyone who had entered and what they had brought with them.
He went home and slept without opening the file.
The third morning he opened the Park family case and found the first sentence.
It came faster than the other two, which was what Maren had said. Not easier — he wrote and deleted twice before he had it — but cleaner, like something that had been waiting for him to show up and had already been assembled.
A child's experience of being known by name is not a supplementary benefit. It is a foundational condition of certain kinds of development that the allocation system does not currently model. When a parent identifies a specific instance of this condition and requests placement accordingly, the system has no framework for evaluating the claim. The absence of a framework is not evidence that the claim is without merit.
He built the recommendation from there. By midmorning he had four pages. He read them back and found one paragraph that was doing too much — trying to argue for the principle and the specific case simultaneously — and split it into two. The principle paragraph went before the assessment. The case-specific paragraph was the assessment.
Determination: Approve placement at Eastbridge Primary School. The petitioner has identified a non-measurable but specific benefit — a teacher's knowledge of the child — and has judged this benefit to outweigh the measurable costs. The bureau cannot verify the benefit's value. The bureau can verify that the petitioner has clearly articulated what they are asking for and why. That articulation, combined with the absence of measurable harm from the alternative placement, is sufficient grounds for approval. The record should note: this case was approved on the basis of a parental assessment of a child's needs. This is appropriate. It should not require justification beyond that.
He read the last two sentences several times.
Then he sent all three recommendations to the folder he had prepared for Chen and wrote her a short note: Three attached. Let me know if anything needs revision before you take them to the ARB.
He closed the files and sat at his desk for a while.
The work of the three days had been — he looked for the word — instructive. Not in the sense of having taught him something he didn't know. More in the sense of having shown him the shape of something he had been carrying without seeing it clearly. He had believed, since Case 7743, that there was a gap between what the system could measure and what was real. He had argued for this in the testimony. But writing the recommendations had required him to be precise about the gap in specific cases, which was different from arguing for it in principle. In each case, the gap had a particular shape. Suki Wen's gap was about cost and benefit being in different registers. Adeyemi's gap was about time and place being undervalued. The Park family's gap was about knowledge of a person being a developmental variable.
Three different shapes. One underlying structure: the system's inability to hold what it could not measure.
He wrote in his notes: The system counts what it can count. The recommendation goes into the record to say: there is more than what is counted.
He looked at the note for a while. Then he added: This is also what Suki Wen was saying, and Adeyemi, and the Park family. Each in their own way, each in the language the form gave them, each saying: there is more than what you are counting. The system heard it as a data error. The record now says it wasn't.
He had spent eight years in the audit queue, and the work had always felt like verification — confirming that what was claimed matched what was documented, that the world matched the record. He understood now that it had always been something else too: a decision, made by the auditor as much as by the system, about what was real and what wasn't. He had spent eight years making that decision one way and then on one Tuesday morning in a year that no longer felt recent, he had made it differently. The three recommendations were that difference, written out in bureau language, filed where it would be read by people who had not been in the room.
Then he put his coat on and went to the Tower.