The migraine lasted four hours.
I timed it.
Not because I expected that to be useful information immediately, but because everything is potentially useful information, and the cost of recording something you don't need is lower than the cost of forgetting something you do. So I lay on the cot in the back room of the office with a cold compress over my eyes and a timer running on my phone and I noted that the pain peaked around the ninety-minute mark, plateaued, and then dropped off sharply in the final thirty minutes like a fever breaking.
Four hours and eleven minutes, total.
The cause: one use of Puppeteer on a courier who'd been loitering outside the building for the third time in two weeks. Simple command. *Walk away and don't come back.* Sustained eye contact for three seconds — I counted — and then the usual sensation, which I can only describe as someone reaching into the base of my skull and closing a fist around something that wasn't meant to be touched.
Then the migraine.
I lay there and did not sleep, because I can't sleep during them, and instead I thought about the math.
One use. One command. Four hours of compromised function. During those four hours I had answered no messages, attended no meetings, reviewed no documents. I had generated no value and made no decisions more complex than *water or no water* and *is that sound outside a threat or not.* If I'd had a time-sensitive job running — and I'd had three of those in the last month — the migraine would have been a critical failure point. An operator who is neutralized for four hours after using his primary tool is not operating. He's recovering. There's a difference.
I had been treating Puppeteer like a weapon I kept in reserve.
That was wrong.
A weapon you can only fire once before putting yourself on the ground is not a weapon. It's a liability with good timing. The only reason it had worked so far was that I'd been careful about when I used it and fortunate about when I hadn't needed to. Careful and fortunate are not a strategy. They're a temporary condition that ends the first time circumstances stop cooperating.
I needed to train it.
---
The literature on quirk development is thin and inconsistent, which is itself useful information about the people who've documented it.
Most of what exists was written by heroes, for heroes, in an institutional context where the primary question is *how do we make this faster and bigger and more impressive in a straight-line fight.* Scaling for combat. That's what gets funded, measured, published. Nobody in the hero pipeline writes papers on how to use a mental-influence quirk with less physiological cost, because the hero pipeline doesn't need that. Heroes operate in short bursts with support structures around them. They can afford four-hour recovery windows. There are protocols for it. There are teammates.
I had teammates in a technical sense. I had a crew, small and functional, and they were reliable within defined parameters.
But I did not have a support structure that could absorb four hours of me being useless. Not yet. Maybe eventually. That was the shape of the problem I was building toward solving.
For now, I needed to reduce the cost.
I started in the library.
Not the corridor archives — those are operational resources, and I try to keep operational and developmental separated when I can, because mixing them creates dependency loops I'd rather not have. The public library on Mizuno Street, the one with the cracked east window that nobody has fixed since the war and the librarian who has decided that her primary job function is to be aggressively invisible. Good woman. I went there on three consecutive mornings and pulled everything I could find on neuroscience written for a non-specialist audience, because I am not a specialist and I am aware of the limits of pretending otherwise.
The core concept I was looking for was *neural fatigue.* The way a muscle breaks down under load and rebuilds stronger. The assumption I was testing was that Puppeteer's post-use cost — the migraine — was analogous to muscular failure after maximum exertion, and that the correct approach was therefore not to avoid exertion but to increase it systematically and allow recovery, so that the baseline capacity expanded over time.
The assumption might be wrong. Quirks are not muscles. The analogy was a working hypothesis, not a conclusion.
But I didn't have a better framework, and no framework is worse than a flawed one, so I took the analogy and started designing around it.
---
The training protocol I settled on was simple enough to be sustainable and precise enough to be measurable.
Every morning, before anything else: one use of Puppeteer. Not in a job context. Not on someone who mattered. On whoever was convenient and low-risk — corridor foot traffic, people who passed through areas I was already monitoring, individuals whose forgetting me or walking a different direction would have no downstream consequences. The command was always minimal. The simplest possible instruction. *Keep walking. Turn left. Look away.*
I logged the time-to-migraine onset. I logged the duration. I logged the intensity on a ten-point scale that I calibrated against my own threshold for functional impairment — five was the point where I could still read but not write, seven was where I couldn't read, nine was horizontal and lights off.
First week: onset immediate, duration four to four and a half hours, intensity averaging eight.
I kept going.
The theory was that if I pushed into the pain window consistently and then rested, the system would adapt. I didn't know if that was true for quirks. I was about to find out.
---
Toga found me on the fourth day with a folder under her arm and the particular expression she gets when she's decided she's going to tell me something I don't want to hear but she's also already prepared for the argument.
"You look terrible," she said.
"I'm aware."
"Are you sick?"
"No."
She looked at me for a moment with those yellow eyes doing that thing where they get very still and very focused, which means she's decided not to push on the current subject and file it for later. That's a habit I've come to appreciate. She has good instincts about when pushing is productive.
"Camie got confirmation on the Saito referral," she said, and put the folder on my desk. "They want a preliminary assessment by end of week. Low complexity, probably. Materials procurement, two suppliers, mid-tier contract value."
"Who's the client contact?"
"Someone called Endo. Came through the Uehara chain."
The Uehara chain. That was legitimate-adjacent referral traffic, which meant the client had probably been told I was a consulting resource rather than a corridor operator. I'd been getting more of those. The legitimization effort from the Moriwaki work was producing exactly the downstream effect I'd designed it to produce, which was satisfying in a quiet, technical way that I didn't have a better word for than *correct.*
"Tell Camie to take the preliminary meeting," I said. "She can run the initial intake. I'll review the output."
"You're not going yourself?"
"Not yet."
Toga's expression shifted fractionally. Not quite concern — she doesn't do concern in any way that looks like concern — but something adjacent to it. Assessment, maybe. "The corridor job last week. The Inoue handoff. You were off."
"I know."
"You made two routing decisions slower than usual. Hatsue noticed."
"I know," I said again. "It's being addressed."
She held the look for another two seconds, then nodded once and picked up two-thirds of the folder, leaving me the intake summary. "I'll brief Camie."
She left without asking what *being addressed* meant.
That's the other thing I appreciate about her. She knows when not to ask.
---
By the end of the second week the numbers had moved.
Not dramatically. This was not a montage. The intensity average had dropped from eight to seven-point-two, which sounds small and is small, but in practical terms meant the difference between *cannot read* and *can read slowly with effort.* The duration had dropped by twenty-two minutes on average. Onset was still immediate.
I noted all of this.
I also noted that I was getting better at the eye contact mechanics independent of the quirk itself — the way you hold a person's gaze without looking like you're trying to hold their gaze, which is a social skill with applications well beyond Puppeteer. Most people break eye contact in about three seconds in a neutral interaction. If I held at two and a half seconds and then let them look away first, they didn't register anything unusual. If I held past four seconds they started to feel it, started to feel looked at, and that tension produced exactly the kind of full attention that Puppeteer needed to function.
So I was also, essentially, training approach mechanics. The delivery system, not just the payload.
That felt right. The work had a shape now. A direction.
I took the Saito folder home and read it in the hour before the migraine onset, because that was still my best cognitive window, and thought about what a mid-tier procurement assessment looked like from a client who'd come through the legitimate referral chain.
Thought about what it meant that the referral chain existed now.
Thought about the ceiling Hatsue had mentioned, in that quiet way she mentions things, the ceiling that might become a door.
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