The first three days were inventory.
Not emotional inventory. I didn't have the patience for that, and the intercostal tears made deep breathing uncomfortable enough that introspection felt like an unnecessary additional cost. I mean literal inventory — lying on the mattress in the corridor with the operations record open, going through every active variable in the system and asking the same question about each one.
*Does this break if I'm out of operation for six weeks?*
Most of it didn't. That was the first useful data point to come out of the whole situation.
Toga was mid-cycle on a standing information contract I'd set up through Ogata's replacement three weeks prior. She didn't need me to complete it. She needed a drop point and a payment schedule, both of which I'd already written into the job parameters. The contract would run to completion without my involvement as long as nothing unexpected interrupted her.
Camie had two pending follow-up contacts with the Meridian buyer's secondary representative. I'd structured those as independent exchanges — she handled the conversation, I reviewed the output. With the review delayed, the timeline shifted, but the relationship didn't break. Camie was good at maintaining surface. It was one of the reasons she was useful.
The broker pipeline — Matsuda's replacement, the courier I'd been developing, the three standing buyers — none of them required my physical presence for normal operations. They required my judgment. Judgment I could still provide horizontally, on a mattress, with two torn intercostals and a painkiller budget I'd already calculated down to the week.
The system held.
That was the second thing I needed to know, and knowing it made the first three days bearable.
By day four I was thinking clearly enough to be angry, which I treated as a sign of recovery.
Not angry at Hando. He'd done exactly what a combat veteran should do when something feels wrong. He'd responded with proportional, contained force and extracted himself cleanly. If anything, I'd collected a data point about how competent mid-tier pros actually behave under ambiguity.
That data point had cost me weeks and a walk-in clinic visit, but it was real.
I was angry at the gap in my own model. The specific failure: I had built a system that kept me outside the room, then walked myself into the room, then failed to account for what happens inside the room when the asset is me.
The gap wasn't theoretical. I'd known, in some abstract sense, that Puppeteer required proximity. Four meters. Sometimes six, if the ambient noise floor was low enough and the target was passive. But I'd been treating proximity as a logistical constraint — something to manage through positioning and timing — rather than a structural vulnerability.
Proximity meant exposure. Exposure meant I could be hit.
I had no answer for being hit.
Months of training All of it — the boxing gym, the karate dojo, the conditioning work I'd been doing for months — and a single palm strike from a mid-tier pro had put me on the floor of a noodle counter.
The training wasn't useless. I understood that. It had built endurance, basic footwork, the ability to absorb minor impacts without folding immediately. What it hadn't built — what it couldn't build in the time I'd given it, and possibly couldn't build at the scale required — was the ability to trade blows with someone whose Quirk enhanced their physical output.
The math was straightforward.
An unenhanced fighter against a Quirk-enhanced one was a different weight class. Not a different skill level. A different weight class. Skill compensated for a great deal. It didn't compensate for a thirty-percent density increase applied to a palm strike from someone who'd been throwing palm strikes for a decade.
I couldn't train my way out of that gap. Not in any realistic timeline.
So the answer wasn't to become someone who could survive direct contact.
The answer was to make direct contact structurally impossible.
I spent days five through eight reading.
The clinic had given me a light mobility restriction — no significant torso rotation, nothing that compressed the rib cage laterally — which made the library more practical than the corridor. Better light. Longer tables. The ability to spread reference materials without the weight of the operations record pressing into the mattress edge.
I went back to the materials I'd worked through during the Puppeteer refinement sessions. Neuroscience. Attention architecture. The pre-conscious processing chain.
This time I wasn't reading to improve the Quirk's efficiency.
I was reading to understand its range ceiling and whether that ceiling was fixed.
The honest answer, after four days of reading, was: probably not fixed, but also not expandable through effort alone.
Range in attention-based Quirks — and I was speculating here, since there was no legitimate academic literature on Puppeteer specifically — likely correlated with signal clarity. The pre-conscious channel I was pushing through was a low-bandwidth pathway. It degraded over distance the way any low-amplitude signal did. More distance meant more noise in the channel, which meant I needed to push harder to achieve the same directional effect, which meant higher cognitive cost, which meant the headache threshold moved forward and the margin for error shrank.
I couldn't extend range without finding a way to increase signal strength or reduce channel noise.
Signal strength was a Quirk property. I didn't know how to alter Quirk properties. That path led toward resources I didn't have and researchers I didn't want knowing my Quirk existed.
Channel noise was an environmental property.
That was interesting.
I thought about it for the rest of day eight.
Channel noise. The concept was intuitive once I framed it correctly.
The pre-conscious attention system processed more information when the target was in a high-stimulus environment — crowds, movement, sound, competing visual inputs. More incoming data meant more parallel processing, which meant my signal had to compete with more noise for the same attention bandwidth. That required more push, more cost, more exposure time.
In a low-stimulus environment — quiet room, isolated setting, minimal competing inputs — the channel was cleaner. My signal reached the target faster, at lower intensity, with less effort on my end.
That wasn't a new observation. I'd noticed it empirically during the early testing sessions. Toga was easiest to nudge when she was waiting, not when she was moving. The gate attendant at the district office had required almost nothing because the space was quiet and the moment was boring.
What I hadn't done was reverse-engineer the implication.
If channel noise determined my effective range, then controlled environments weren't just safer for other reasons — they were mechanically superior for Puppeteer. I'd been treating environment as a variable I worked around. It was a variable I could engineer.
A target alone in a quiet room, at the right moment of cognitive load — tired, mildly hungry, slightly bored — was a target I could potentially reach from ten meters instead of four.
I didn't have proof of that. I had a model.
But the model had a different shape than the one that had put me on the floor of the noodle counter, and that was something.
I wrote it down.
*Range is not fixed. Range is environmental. Control the environment, expand the ceiling.*
Below it I wrote:
*Never approach a target in their chosen space again.*
Toga came by on day nine.
I hadn't asked her to. She arrived with food she'd purchased with her own money, which was unusual enough that I noted it without commenting, and sat on the floor against the opposite wall with the particular quality of stillness she had when she was assessing something.
"You look bad," she said.
"Intercostal tears. Six-week load restriction."
"I know what happened."
I looked at her.
"Not the specifics," she said. "I know you tried something and it went wrong. You've got the face you have when you're redesigning something."
I considered whether to confirm that. She'd get the information eventually through operational channels — the job delay would be visible to her even if I didn't explain it. Withholding the specifics without reason was just pride, and pride wasn't a variable I budgeted for.
"I miscalculated a proximity contact," I said. "The target was a trained combatant. The model didn't account for combat-trained attention management."
She was quiet for a moment.
"He hit you."
"Once."
"Did you fall?"
There wasn't any point in lying about something she'd eventually reconstruct anyway. "Yes."
She didn't say anything else immediately. I watched her process it. Toga's emotional responses were visible if you knew what to look for — not in her face, which she controlled better than most people assumed, but in the timing of her silences.
This one ran a beat longer than usual.
"The job's still running," she said finally. "The drop is Tuesday. I don't need anything from you before then."
"I know. I checked the contract parameters."
"I wasn't reporting." She paused. "I'm telling you it's handled."
I understood the distinction she was drawing. I didn't have a useful response to it, so I didn't offer one.
She ate half the food she'd brought, left the other half, and went without discussing payment terms or the next job or anything operational at all.
I ate the food. It was better than what I would have purchased myself.
I noted that too, in the separate part of the record I kept for variables I hadn't fully categorized yet.
By the end of the second week, I had a restructured framework on paper.
It had four components.
The first was environment control. No more approaching targets in open public spaces on their routine. Every Puppeteer application going forward required a setting I'd either chosen or could influence — quieter, lower stimulus, ideally with some element of isolation I'd engineered in advance.
The second was layered distance. I drew a revised range model: confirmed effective range in high-noise environments, estimated effective range in controlled environments, and a hard exclusion zone — anything requiring me to get within two meters of an unknown target was automatically redesigned or reassigned.
The third was exit architecture. Every plan required at minimum two exit routes I hadn't used before. Not for emergencies. As defaults. The noodle counter had one open side and a railing. I'd sat near the railing. That was the kind of spatial thinking that got people hit.
The fourth was the one that was harder to write down, because it touched something structural about the whole system.
*Kuro cannot be the proximate instrument.*
Not just the operator who stayed outside. Not just the one who planned. If a job required someone within Puppeteer range of a target who might respond with force, that someone had to either be an asset I could afford to lose or a situation I'd engineered well enough that force was never a realistic outcome.
I wasn't there yet. I didn't have an asset with that profile. Toga was valuable and I couldn't afford to lose her. Camie was newer and I couldn't afford to burn the investment. I had no one else but myself
That meant the fourth component was also a hiring criterion.
I needed, eventually, someone between me and physical risk.
I didn't have a name for that role yet. I wrote it as a placeholder.
*Buffer asset. Physical profile preferred. Timeline: open.*
Next day I went back to the boxing gym.
Not to train. The restriction was still active, and I wasn't interested in re-tearing anything for the sake of routine. I went because I'd paid the monthly fee and because showing up, even modified, was better data for the gym's assessment of my consistency than not showing up at all.
I told the trainer I'd taken a bad impact in a street incident and needed modified sessions. He didn't ask for details. I ran footwork drills for forty minutes, nothing above the waist, and watched the other members work.
There was a man in the far corner I'd seen before but never paid attention to.
Late twenties. Dense build — not large, but assembled the way people are assembled when they've been absorbing impact for years rather than just applying it. He had the footwork of someone with formal training and the hands of someone who'd fought outside the gym. The combination was specific. Not sport. Not street. Something in between, the way certain people from certain backgrounds moved.
He wasn't remarkable. That was what made him interesting.
He wasn't trying to be seen. He was just there, working, competent and quiet about it.
I watched him for the rest of the session.
Didn't speak to him. Didn't adjust my position to signal interest. Just observed, the way you observe a load-bearing wall before you decide whether the building can be modified around it.
On the way out, I confirmed his approximate schedule with the front desk through a question framed as my own scheduling conflict.
Tuesdays, Thursdays, and occasional Saturdays.
I wrote it in the buffer asset section of the restructured framework.
*Possible. Observe further.*
That evening I sat in the corridor with the operations record and looked at everything I'd written across sixteen days.
The system was different now. Not larger — not yet. But differently shaped. The failure had removed a structural assumption I'd been building on without knowing it, and removing it had forced a rebuild that was, in several ways, more honest than the original.
I hadn't accounted for what happened when the operator became the point of contact.
Now I had.
I closed the record.