The second bank job paid better than the first.
Not because we were greedier. The payout structure was the same—Kuroda's flat fee for clean extraction, no bonus for speed, no penalty for complications provided the complications didn't reach him. The difference was that the first job had cost us more than it needed to, and I had spent the two weeks between them studying the failure points with the same attention I used to study everything.
Systematically.
Without ego.
Treating my own decisions as objects I could pick up and turn over in bad light until I understood their shape.
That process took most of the first week. Not because the failures were complex—they weren't—but because I had to be certain I was seeing the decisions themselves rather than the outcomes. Outcomes lie. A correct decision can produce a bad result through variance. An incorrect decision can produce a good result through luck. If you optimize for outcomes, you're optimizing for luck.
I was not interested in luck.
The first job had three inefficiencies. I found the first one on the second day of review, sitting in the Kanda Street office with the building's old heating system clicking in the walls and Nori's interference logs spread across the desk in front of me. Riku had held position forty seconds longer than necessary. The reason was visible in the comms transcript: I had given him a conditional exit signal rather than a clear one. "Move when you have a clean window" instead of "Move at mark." The conditional had forced him to evaluate, and evaluation took time, and forty seconds of unnecessary stationary exposure was the result.
My error. Not his.
I noted it and kept looking.
The second inefficiency took longer to isolate because it required reconstructing Nori's prep timeline against the actual interference environment we encountered. His calibration had been set for worst-case signal density—a professional precaution that had cost us preparation time on a margin we didn't end up needing. The margin had been reasonable. The cost had been real. Whether the tradeoff was correct depended on variables I hadn't given him, because I hadn't specified what kind of margin I wanted him to optimize for.
Another error.
The third inefficiency was structural. The extraction routing had included an unnecessary relay hop that added transfer delay without adding meaningful security. I had approved the routing without questioning it because it matched a template I'd used before, and I had been focused on other elements.
Three errors. None catastrophic. All correctable.
The second job ran in eleven minutes. Clean extraction. Clean egress. No unnecessary contact. Nori shaved four minutes off his calibration window and the margin held—narrower than before, but still within tolerance. Riku exited exactly on signal. The routing was direct.
Hatsue said afterward, in the van, with the city sliding past the windows and the extraction package secured between her feet, "That felt different."
She wasn't looking at me when she said it. She was watching the street, doing her job, maintaining perimeter awareness even in withdrawal. But her voice carried something I filed for later examination.
I said, "It was supposed to."
---
By the end of the month the team had a rhythm.
The word came to me while reviewing Nori's updated interference protocols at three in the morning, alone in the office. Rhythm. Not comfort—I was careful not to let it become comfort. Rhythm is functional. It's the elimination of unnecessary friction, the development of shared expectation that allows faster coordination. Comfort is what happens when people stop checking their work because the last several times it was fine.
Those are different things.
I said so once, directly, about ten days in. The four of us were in the main room of the Kanda office, mid-afternoon light coming through the frosted windows, and I had finished walking through the next job's sequence. Daigo asked a clarifying question about his comms protocol. Nori nodded along. And I paused—deliberately, letting the silence extend just long enough to shift the room's attention—and said: "This works because we check it. The moment we stop checking it, it stops working. Not gradually. Immediately."
No one argued. Riku met my eyes for a moment and then looked away, not in avoidance but in acknowledgment.
Nori tested the boundary later that same week.
He came to a job prep meeting with half his interference specs completed. He had made a judgment call—the target location was a known profile, one we'd worked twice before, and he had assumed the full range wouldn't be required. The assumption was correct. The problem wasn't the judgment. The problem was that he had made it without flagging it.
I found out when I asked to review his completed prep and he showed me what he had instead of what I expected.
I sat with that for a moment. Long enough that he shifted in his chair.
Then I said: "If you make a call, I need to know you made it before we're in the building. Not after."
My voice was level. I wasn't angry. I was calibrating.
He said, "I didn't think it mattered since—"
"It matters because I'm operating on incomplete information about my own operation. The call itself was fine. The silence about the call is what I need to not happen again."
He understood. The tension left his shoulders—not all at once, but gradually, like air escaping something pressurized. He didn't push back. That told me something useful about him. He wanted to be seen as capable more than he wanted to be right, which meant his ego was manageable if you gave him a path to competence that didn't require admitting he had been wrong.
That was more than I could say for some people I'd worked with.
---
The love scam was Daigo's domain and he ran it with a precision I hadn't expected.
I had assumed, when Kuroda first described his role, that Daigo's function was primarily social—the face, the contact, the person who could talk to marks without triggering their suspicion. That assumption was incorrect. Daigo had built four distinct personas, each with a complete history reconstructed from pre-war social media archives, consistent communication patterns, documented grief that matched verifiable public events, plausible backstory elements that would hold up under casual investigation.
He ran all four simultaneously.
I watched him work one evening from across the office. He had three windows open on his display, each a different conversation with a different mark, and he rotated between them with the same unhurried attention I used when cross-referencing job variables. Response times were calibrated to feel natural—never immediate, never so delayed that interest waned. He wrote like someone who had spent years studying how people communicate when they're lonely.
Hatsue tracked the targets.
She had a gift for pattern recognition that went beyond her quirk. She could look at a communication profile for forty minutes and tell you whether the person was lonely in a way that was recoverable or lonely in a way that had calcified into desperation. She showed me the distinction once, walking me through a set of linguistic markers I wouldn't have noticed on my own: the frequency of self-reference, the way apologies were structured, the presence or absence of future-oriented language.
Recoverable loneliness still imagined a future where the loneliness ended. Calcified loneliness had stopped imagining.
We only ran the long-contact personas at the second type.
I had made that a rule. Not out of mercy. Out of efficiency.
I spent an evening documenting the logic in my records, alone in the office after the others had left. Desperate loneliness responds faster and deeper. The recovery timeline is shorter because there's less cognitive resistance to the initial contact. A person in recoverable loneliness might eventually talk themselves back toward skepticism—they still have enough internal structure to question why someone is suddenly interested in them. A person in calcified loneliness needs the contact too much to let it go, which means they won't examine it closely, which means the extraction is cleaner.
I wrote all of this down.
Then I closed the file and moved to the next problem.
---
The phishing infrastructure had scaled.
Hatsue had cross-referenced three separate data sets—reconstruction contractor registrations, emergency procurement portals, and a leaked internal directory from a mid-tier construction firm that Nori had acquired through a previous job. The directory was six months old but still functionally accurate for most entries. The cross-reference had identified a cluster of forty-seven targets in similar credential vulnerability windows: people who had recently changed positions, people whose contact information had been updated in one system but not another, people whose verification chains had gaps created by wartime record loss.
Forty-seven was too many to run manually.
I spent a day redesigning the approach. Nori had built a shell account infrastructure for previous corporate sabotage work—nothing sophisticated by the standards of people with real technical resources, but clean and consistent and hard to trace without dedicated effort. I adapted it to a semi-automated pipeline. Initial contact through a reconstruction procurement portal impersonation. Credential capture through a cloned verification page. The infrastructure handled the first layer; human review handled anything that required judgment.
The pipeline converted at thirty-one percent.
Fourteen new accounts. Lower average value per account than the manual work—the automation couldn't discriminate between high-value and low-value targets the way Hatsue's analysis could. But the time cost per extraction had dropped to almost nothing once the infrastructure was running.
I noted the ratio. I thought about what that ratio would look like at double the target pool.
Then I thought about what it would look like at ten times.
I let that thought sit for approximately four seconds.
Then I closed it.
Not because it wasn't useful. Because I was not ready to operate at ten times the target pool, and thinking about scale I couldn't yet support was a different form of the same mistake that had put me on the floor of a hospital two months ago. Ambition was a tool. Tools required calibration. You don't use a welding torch when you need a screwdriver just because you happen to be holding a welding torch and imagining what you could build with it.
I went back to the fourteen accounts and began the extraction sequencing.
---
Riku found us a problem on a Wednesday.
He came in from his street-level monitoring work in the early afternoon. Part of his role was physical presence in three neighborhoods—not surveillance, not yet, just presence. Pattern familiarity. The kind of background visibility that meant when we needed low-profile observation, he had existing cover. Shopkeepers recognized him. Delivery drivers nodded. He was part of the street's texture.
He waited until Hatsue and Nori had left for a supply run before he spoke.
"There was a man asking about the Kanda Street office."
The office was quiet. Just the heating system and the distant sound of traffic from the main road.
"What kind of asking," I said.
"Casual." Riku's voice was careful—not nervous, but precise, like he was trying to give me exactly what he had observed without interpretation. "Like someone who already knows the answer and is checking whether other people know it too."
I kept my expression neutral. "Describe him."
Mid-forties. Reconstruction-era jacket with the sleeve patching style common to the northern ward displacement camps—a specific diagonal stitch pattern that was distinctive if you knew to look for it. Calm in the way people are calm when they have already made their decisions. He had asked two shopkeepers, both within a three-block radius of the office, and then not appeared again.
"Did he ask about any of us by name."
"No. Just the building. Whether there was an office on the second floor that saw regular traffic."
I thanked Riku and sent him home.
Then I sat alone for two hours.
The list of people who knew about the Kanda office was short by design. Kuroda. Puppimil. The four team members. Fujisawa the building owner, who had been discreet because discretion was his business model and he understood that his continued income depended on it.
The description didn't match Kuroda's people—he used a different type, younger and harder to read. Puppimil wouldn't ask shopkeepers; she had better methods and no reason to waste time on low-yield approaches. The team members had no reason to probe their own location.
That left two categories.
Someone who had followed a thread from a previous job. Someone who had noticed a pattern we hadn't concealed well enough, who was working backward from effect to cause, who had enough patience to ask questions casually and enough discipline to stop before they drew attention.
Or someone who had been sent.
I didn't know which yet.
The distinction mattered. A thread-follower was a problem I could solve by identifying and cutting the thread. Someone who had been sent meant there was a sender, and the sender was the real problem, and the man in the reconstruction jacket was just the visible edge of something I couldn't yet see.
I flagged the office as a soft compromise in my internal classification system and began identifying a secondary location.
---
The week after that I went back to the Yaoyorozu file.
Not because something had changed. Because I had promised myself I would return to it when I had enough floor to stand on, and I had been checking that condition every week since I made the note. The check was simple: Did I have legitimate income, or income that functioned as legitimate? Was the team stable? Did the operation have enough redundancy that I could afford to spend cognitive resources on something that was not immediately productive?
This week the answer was different.
I waited until evening. The others had gone. The office was mine. I pulled the file from its encrypted location and opened it on the desk display, and for a moment I just looked at the structure of what I had collected.
---
Yaoyorozu Momo.
I had three weeks of additional passive accumulation since I'd first flagged her. News archives. Public board minutes. Social mentions that survived the war's data degradation. Location patterns reconstructed from event attendance records and public appearance documentation.
The picture was sharper now but still incomplete.
She was present at four separate materials allocation consultations in six weeks. Unpaid advisory role, as noted in the initial file. But something else had become visible: each meeting was preceded by a materials assessment report that bore the methodological fingerprints of someone with direct Creation quirk application. The analyses weren't just technical—they showed an understanding of material properties that could only come from someone who had actually created the materials in question, who had felt the molecular structure form under their direction and understood what that structure could and couldn't tolerate.
She was doing the analytical work that informed the contracts she was then nominally consulting on.
She was also being kept at arm's length from the decisions themselves.
I turned that observation over for a long time.
It suggested whoever was using her was doing so carefully. They wanted her capability without her influence. They understood that her analytical work was valuable—valuable enough to structure meetings around reports that bore her imprint but not her name. But they also understood that her influence, if she chose to exercise it, would be a complication for whatever they were actually trying to achieve.
She had leverage she wasn't using.
I spent the next hour trying to understand why.
The most likely answer was structural rather than personal. She had come from a hero framework—a culture that trained people to act within designated structures, defer to institutional authority, contribute without demanding recognition. The hero licensing system had been that structure. It had given her a clear role, clear expectations, clear metrics for success. The war had dissolved it. She had moved into the reconstruction framework and, from what I could see, replicated the same pattern: contribute, defer, trust the institution.
She was twenty-one.
She had probably never had a reason to question whether the institutions she trusted were worth trusting.
I sat with that for a while. Not with pity—pity was a distortion, an emotional response that would cloud my analysis. With the specific quality of attention I gave to structural vulnerabilities in a job plan. She was a load-bearing element in a system she didn't fully understand. She was providing value that others were capturing. And she appeared to accept this arrangement as natural.
That created a force gradient across her position. The difference between what she contributed and what she received in return—in influence, in recognition, in control over the outcomes her work enabled—was a form of potential energy.
Force gradients could be redirected.
But not quickly. Not clumsily. The wrong approach would close the window permanently. If I approached her directly and she rejected the approach, she would become defensive. If I tried to show her the pattern and she wasn't ready to see it, she would rationalize it away. And once someone has rationalized away a truth, presenting it again is twice as hard.
I needed to understand what she wanted before I could understand what she would respond to.
Not what she thought she wanted. The things people think they want are usually just the things they've been taught are acceptable to want. Heroic recognition. Institutional approval. A seat at a table whose shape they didn't choose.
I needed what she actually wanted. The thing underneath the hero framework and the family foundation and the consultancy work. The thing that the war had revealed or buried or both. The thing that made her continue doing analytical work that others took credit for, week after week, without visible resentment.
That required closer information.
Not contact. Not yet.
Observation first. Always observation first.
I started a new document and labeled it YMM — Structural Assessment, Phase One.
Then I wrote, at the top of the first page, a single question:
What does she think she's building?
I saved the file and sat looking at the question for a long moment. It was the right question. I was fairly sure of that. But I didn't have an answer yet, and I wouldn't for some time, and in the meantime there was other work to do.
I closed the document.
---
The secondary office location came through by Friday.
Fujisawa had a contact in Suidobashi—a basement unit under a legitimate records storage company, shared entrance, the kind of arrangement where foot traffic was expected and unremarkable. The company's business provided natural cover. People carried boxes in and out. No one asked questions because the whole point of records storage is that you don't ask questions about what's in the boxes.
I moved the core operation documentation that weekend. Financial records. Job files. Contact protocols. Anything that would be difficult to reconstruct from memory if the Kanda office was compromised. I left enough in Kanda to maintain the appearance of functionality—a meeting point, not a nerve center.
Fujisawa had been discreet but he was a single point of failure. I had been aware of that since the beginning. The man in the reconstruction jacket had simply reminded me that awareness and mitigation are different things.
The man didn't appear again.
I had Riku maintain the monitoring pattern for another week. Nothing. The shopkeepers hadn't seen him. The street was its usual self.
That worried me slightly more than if he had.
People who stop asking questions usually already found their answer.
I noted it. Filed it. Kept one additional layer of passive observation on the street outside Kanda—nothing active, nothing that would draw attention, just a slight adjustment to Riku's existing pattern that would flag anything unusual.
And I went back to reading Yaoyorozu Foundation public filings in the new basement office.
The ventilation system from the records storage company hummed on the other side of the wall—a constant low sound that took some adjustment but eventually faded into background. The lighting was worse than Kanda. The space was smaller. But it was secure, or as secure as anything could be, and that was what mattered.
I had the Yaoyorozu materials spread across the display. Public board meeting minutes from three months ago. A procurement announcement that mentioned her name in passing. A photo from a reconstruction site visit, her face half-turned away from the camera, her attention on something outside the frame.
The question was still there at the top of my document. Unanswered.
What does she think she's building?
Visit My Patreon and read upto 30+ Advanced chapters in advance to support the story
My Pat reon:ht tps://patreon.com/eyestapler
Thanks for reading