Uehara's active review status came in on a Tuesday.
Seven business days, which was the outer edge of the window I'd given him. I'd been checking the relay each morning with the specific quality of attention you give to something you're not supposed to be anxious about and are slightly anxious about anyway. Not because I doubted the mechanism—the reconstruction audit division's intake process was documented and I'd used it correctly. Because I'd told Hoshino three weeks and the three weeks had a foundation, and foundations that don't set properly on day seven don't get stronger on day twenty-one.
The confirmation arrived at 8 AM. Reference number assigned, complaint classified as active irregularity review, assigned examiner on record.
I forwarded the reference number to Hoshino with one line:
*Active. Examiner assigned. Clock is running.*
He replied in eleven minutes, which was the fastest he'd ever responded to anything:
*Understood. I'll hold to the window. Editor has been managed.*
I set the phone down and made coffee and allowed myself exactly thirty seconds of something that wasn't quite satisfaction but was adjacent to it—the particular internal register of a load-bearing piece sliding into position correctly. Then I opened the Aozora file and got back to work.
---
Hatsue had delivered the taxonomy Friday evening, left on my desk with the same quiet efficiency as the first draft, no ceremony. But the full version was different in kind from the draft—denser, more confident in its structural claims, with a section at the back she hadn't mentioned that cross-referenced the eleven fraud pattern types against publicly available regulatory filing data to produce a rough frequency estimate for each pattern type across the full recovery zone.
She'd done epidemiology on post-war procurement fraud.
With the resources available to someone working in a basement.
I'd read it twice over the weekend and marked three sections with questions, not because they were wrong but because I wanted to understand her reasoning at the specific points where she'd made inferential leaps rather than following direct evidence chains. When she came in Monday I asked her about all three.
Her answers were good. Better than good—they were the answers of someone who had thought about the objections before I raised them and had a structural response prepared rather than a defensive one.
I told her to add the reasoning to the document explicitly, so the inferential steps were visible rather than buried.
"Makes it look less certain," she said.
"Makes it look more honest," I said. "Those aren't the same thing, and the audience we're eventually building this for will value the honesty."
She wrote something in the notebook.
Now, Tuesday morning, I was working the taxonomy against Sera Michiko's aggregate summary, building the formal mapping that was the first deliverable in our Aozora engagement. It was detailed work—matching Hatsue's abstracted pattern signatures against the specific offset distributions Sera had documented across her firm's twenty-two months of site assessments, then translating the alignment into language that an audit division examiner would receive as professional methodology rather than information brokerage.
The language mattered as much as the content. An examiner reading a submission from a reconstruction consultancy needed to see the work of a reconstruction consultancy—precise, physical, grounded in site observation—rather than the work of whatever I actually was.
I was good at writing the work of whatever other thing was needed.
---
The morning ran clean until Riku came in at eleven with a look I recognized.
Riku's operational communication was mostly tonal. He had a look for routine intake, a look for complications requiring immediate attention, and a look for things he'd encountered that didn't fit either category and were therefore either irrelevant or important and he wasn't sure which.
He had the third look.
"What," I said.
He set a single printed page on the desk. Not a message—a capture. Something he'd pulled from the corridor monitoring system Daigo maintained, flagging unusual activity around our known relay channels and location signatures.
Someone had run a search query through one of the corridor's information brokerage networks. The query was specific: it referenced Shinkō Data Services by name, combined with search parameters for *reconstruction contract dispute consulting* and *HPSC intake facilitation.*
Not a client inquiry. Clients came through referral chains and used vague language. This was a direct name-and-function search, which meant whoever ran it already knew what we were and was verifying scope rather than discovering us.
I read the capture twice.
"When," I said.
"Yesterday afternoon. Between 2 and 4 PM."
During my working session with the Aozora mapping. While I'd been building the cross-reference document with the window closed and the secondary monitor running pattern data.
"Source trace?"
"Daigo got a partial. The query ran through a corridor node in the Sumida access layer. That node has two regular operators—one is a logistics information broker we've dealt with before, low-tier, not a threat. The other—" He paused. "Daigo thinks it's new. Not a registered corridor identity. Someone using the infrastructure without a standing position in it."
Someone who knew how to use the corridor without being of the corridor.
That was a specific skill set.
I looked at the capture for another moment. Then I set it down and turned back to my monitor.
"Tell Daigo I want full traffic analysis on that Sumida node for the last thirty days," I said. "Not just our signature—everything that moved through it. And I want the logistics broker's recent query history if he can get it without flagging the pull."
"Timeline?"
"Before end of day."
Riku took the page and left.
I looked at the Aozora mapping on my screen. Returned to it. Kept working.
---
The thing about a late-stage complication is the temptation to treat it as a crisis. The instinct is to stop everything and resolve it before moving forward, because an unresolved threat variable adjacent to an operation in its most sensitive phase feels like a structural risk that supersedes everything else.
The instinct is usually wrong.
A late-stage complication is exactly as serious as it actually is, and not more. The correct response is assessment, not suspension. You keep the operation moving because the operation is the thing that matters, and you assess the complication in parallel, and you don't let the uncertainty of an unresolved variable convince you to slow down in a window where slowing down is the actual risk.
I had twenty days left on Hoshino's clock. The Aozora submission needed to be filed before the audit examiner's initial review period closed. Uehara's complaint was in active status and needed no action from me, but I needed to be available for the second tranche trigger when it came.
Three threads, converging. Twenty days.
Whoever had run that query could wait until end of day.
I kept working.
---
Daigo's traffic analysis arrived at 6 PM.
He'd pulled thirty days of Sumida node query logs and run them against a baseline of normal corridor traffic patterns. The result was cleaner than I expected: the unregistered operator had used the node on four separate occasions over the past three weeks, each time running searches that were specific and function-targeted rather than exploratory.
First query, eighteen days ago: *post-war information broker network Musutafu underground.*
Second query, twelve days ago: *reconstruction procurement fraud documentation sources.*
Third query, six days ago: *HPSC contract review intake facilitation corridor.*
Fourth query, yesterday: *Shinkō Data Services reconstruction consulting.*
I read the progression twice.
It was a research arc. Whoever this was had started broad—locating the underground information network generally—and had narrowed over three weeks toward our specific operation. Not randomly. Systematically. Each query was informed by the answer to the previous one, which meant they were following a real evidence chain rather than guessing.
And the chain had led them, in four steps, from the general existence of the corridor to our name.
That was not a random walk. That was someone who had started with a piece of information about what we'd done and had worked backward toward who we were.
The piece they'd started with was almost certainly the Shiro-Gale outcome. That was the most visible of our completed jobs—the one where a legitimate agency had filed a formal complaint with the HPSC and received active review status through what looked, from outside, like unusually effective facilitation. It had been clean enough not to generate corridor noise. It had apparently not been clean enough to avoid generating attention in a space I hadn't been monitoring.
The question was who was doing the monitoring and why.
---
I brought Hatsue and Riku together at 7 PM. Not Daigo—Daigo's value was technical, and this was a judgment conversation.
I laid out the query progression on the table between us.
Riku's first instinct was containment—pull back from the Sumida node, change relay protocols, increase distance from any channel the unregistered operator might have mapped.
"That's the right move if we're the target," I said. "What if we're not the target?"
He looked at the queries again.
Hatsue said it before he reached it: "They're following the same architecture we are."
I nodded. "The second and third queries aren't about us specifically. They're about the fraud documentation and the HPSC intake facilitation. We're one data point in a larger research pattern, not the subject of it."
"Then what are they looking for," Riku said.
"The same thing Hoshino is looking for," I said. "Or something adjacent to it. Someone is mapping the space around the Moriwaki architecture from a corridor-adjacent angle, and they found us because we've been operating in the same space."
Hatsue was already writing. "That makes them either a competitor or a collaborator."
"Or neither," I said. "Could be someone Moriwaki's principals have tasked with identifying who's pulling threads. Could be an independent operator with their own interest in the architecture. Could be someone we don't have a category for yet."
"How do we find out which," Riku said.
I'd been thinking about that all afternoon.
The standard move was surveillance—try to identify the operator through the node, map their other activities, build a picture from behavior. Slow. Uncertain. The kind of approach that worked over weeks and might produce nothing conclusive.
The move I was actually considering was different.
"We answer the query," I said.
Both of them looked at me.
"The query found us by name," I said. "That means whoever ran it now knows we exist and has a general picture of what we do. If they want to know more, they'll reach out—through the corridor, through a referral chain, through some channel that creates the appearance of normal intake. If they don't reach out, it means we're not what they're looking for and the query was incidental."
"And if they do reach out," Hatsue said.
"Then we treat it like any other intake. Learn what they want. Assess the threat level from inside the conversation rather than from outside it."
"That's a significant risk," Riku said. "If this is someone from Moriwaki's side—"
"Then the conversation tells us that immediately and we know something more important than what we know now." I looked at him. "The risk of not knowing who this is over the next twenty days is higher than the risk of a controlled intake meeting."
He absorbed that. Didn't push back further.
Hatsue had stopped writing. She was looking at me with the expression she got when she was deciding whether she agreed.
"You think it's not adversarial," she said.
"I think the query progression reads like research, not targeting. Adversarial operators who know enough to find us don't spend three weeks narrowing a search arc—they send a message." I paused. "But I could be wrong, and I'm not ruling it out."
"So we wait for contact."
"We wait for contact. And while we wait, we keep moving."
---
I worked until nearly ten.
The Aozora mapping was close to complete—another half day, maybe less. Hatsue's taxonomy had made it faster than I'd expected, the pattern alignment coming through clearly enough that the formal documentation almost wrote itself once the structural matches were established.
Before I left, I drafted the submission cover letter for Sera Michiko's review. The language was careful, calibrated to land on an audit examiner's desk as the work of a methodologically rigorous consultancy that had discovered, through routine quality review of its own accumulated assessments, a systematic discrepancy it was obligated to report.
That was true. All of it was true.
The fact that I had helped Sera Michiko understand what her data meant and how to position it didn't make the data or the discrepancy any less real. The architecture existed whether or not I was the one who'd helped her see it
Visit My Patreon and read upto 30+ Advanced chapters in advance to support the story
My Pat reon:ht tps://patr eon.com/eyestapler
Thanks for reading