Matsuda Touta noticed the pattern because nobody else was looking for one.
That was the first mistake. His.
The Task Force conference room smelled faintly of old coffee and overheated electronics, the kind of neutral misery that made hours blur together. Whiteboards were packed with timelines, victim counts, and red-string logic that had been revised so many times it no longer resembled its earlier versions. L sat hunched in his chair, knees up, eyes on a monitor. Soichiro Yagami stood rigid near the window, arms crossed, gaze fixed somewhere beyond the glass. The others were silent, the way people get when they think silence equals seriousness.
Matsuda was flipping through printouts. He wasn't supposed to be. The documents had already been reviewed, summarized, abstracted, and reduced to bullet points. His job was to listen, nod, and occasionally ask something that could be answered in one sentence.
Instead, he kept flipping.
Execution times. Broadcast delays. Medical examiner reports. Local news timestamps. International rebroadcast edits. All things they had already seen. All things they had already decided were noise.
Except the noise lined up.
He didn't realize it at first. It wasn't a revelation. No cinematic click. Just irritation. A vague itch at the back of his brain, like when a song repeats the wrong lyric and you can't stop hearing it.
He marked a page with his thumb.
Another with a pen.
Then he stacked three reports side by side.
The heart attack cases had always been the anchor. Same cause of death. Same absence of physical interaction. Same message. Kira's signature. That part was stable. What wasn't stable was _when_ the deaths occurred relative to observation.
Matsuda frowned.
He checked the timestamps again.
One case in Osaka. One in Fukuoka. One overseas. Different days. Different suspects. Different levels of media exposure.
Same delay.
Not identical. Never identical. That would have been too easy to dismiss. But close enough to rhyme.
"Um," Matsuda said.
No one responded.
He tried again, louder. "Uh, Chief?"
Soichiro turned slightly, not fully. "Yes, Matsuda?"
"I was just, uh, looking at the timing again."
L didn't move.
"So were we," Aizawa said flatly.
Matsuda swallowed. "Right. I know. I mean, I was looking at it in a different way."
That earned him a glance. Not interest. Tolerance.
"Explain," L said, eyes still on the screen.
Matsuda stood up before he could lose his nerve. That was mistake number two. Standing implied confidence. He didn't have any to spare.
He walked to the board and pointed at three cases. "These deaths happened after the suspects were discussed internally. Not publicly. Just… discussed."
Aizawa frowned. "That's coincidence."
"Yes," Matsuda said quickly. "I mean, probably. But the gap is consistent. There's always a delay after—after we look at them closely."
Silence.
"So?" someone asked.
"So," Matsuda continued, words speeding up, "what if it's not just about being seen on TV or named in public? What if attention itself matters?"
That got L's attention. Slightly. His head tilted.
"Define attention," L said.
Matsuda hesitated. "Awareness. Focus. Observation."
Aizawa exhaled sharply. "You're saying Kira knows when we look at someone?"
"No," Matsuda said. "I'm saying maybe it doesn't matter if he knows."
That did it. L turned fully now, dark eyes fixed on Matsuda like pins.
"Continue."
Matsuda's hands were shaking. He clasped them together. "I don't think Kira reacts to information alone. I think the act of analysis changes something. Like… like measuring a system."
No one laughed. That was worse.
"You're proposing an observer effect," L said.
Matsuda nodded, relieved someone else had said it first. "Yes. Not exactly physics, but… similar."
Soichiro shook his head slowly. "Matsuda, that's speculation."
"I know," Matsuda said. "I know. But every time we narrow our focus, the suspect pool collapses. And someone dies. Not the same person every time. But someone."
Aizawa crossed his arms. "That's confirmation bias."
"It might be," Matsuda agreed immediately. Too immediately. "But I checked cases where we didn't pursue a lead deeply. No deaths. At least, not right away."
L was quiet again.
Finally, he said, "Correlation does not imply causation."
"Yes," Matsuda said. "That's what I'm saying. We can't prove causation. But what if the correlation is the mechanism?"
No one spoke.
The room felt smaller. Like the walls had leaned in a few centimeters while no one was watching.
"So what are you suggesting?" Soichiro asked.
Matsuda opened his mouth, then closed it. He hadn't thought that far. He'd just noticed. Noticed and refused to stop noticing.
"I don't know," he admitted. "Maybe we should… change how we observe. Or when."
Aizawa scoffed. "We're not going to stop investigating because you feel weird about timing."
"I didn't say stop," Matsuda said. "Just… be careful."
L looked back at his screen. "Your hypothesis is unprovable."
"Yes," Matsuda said.
"And unfalsifiable with our current constraints," L added.
"Yes."
"Then it is useless," Aizawa said.
L raised a finger. "Not entirely."
Everyone froze.
L continued, "An unprovable hypothesis can still generate testable behavior. If observation alters outcomes, then controlled observation should produce measurable distortion."
Matsuda's heart jumped. "So you think—"
"I think," L interrupted, "that you noticed something because you are not optimized for this task. That makes you unreliable. It also makes you valuable."
Aizawa stared. "You're not seriously—"
"We will not act on this yet," L said. "But we will log it."
Matsuda exhaled, relief flooding through him.
Then L added, almost casually, "And from now on, Matsuda Touta will not be informed when a suspect enters deep analysis."
The relief vanished.
"What?" Matsuda said.
"If your awareness is a variable," L said, "we remove it."
Soichiro frowned. "Is that necessary?"
"Yes," L replied. "If Matsuda is correct, then doubting him is part of the system."
Matsuda opened his mouth to protest, then stopped. Something about that sentence felt wrong. Not incorrect. Wrong.
"I understand," Matsuda said quietly.
The meeting adjourned shortly after. No conclusions. No decisions. Just updated protocols and another layer of silence.
As Matsuda gathered his papers, he noticed something else.
One of the cases he'd marked earlier now had a revised timestamp. Only by a few minutes. Probably a clerical correction.
Probably.
He stared at it anyway.
Because he couldn't shake the feeling that the system had noticed him back.
Matsuda was excluded efficiently.
Not dramatically. No locked doors, no raised voices. Just subtle friction added to his day. Files arrived late. Briefings summarized instead of shared. Screens angled away when he entered the room. Conversations paused, then resumed at a lower volume that carried no usable information.
It was clean. Professional. Surgical.
Which meant it worked.
He sat at his desk with a legal pad he wasn't supposed to need anymore. The Task Force had gone almost fully digital months ago, but paper had one advantage. It didn't log access times. It didn't notify anyone when you stared at it too long.
He started drawing charts.
Not timelines. Those were contaminated. Everyone used timelines. He needed something orthogonal. Something stupid-looking enough to survive scrutiny.
So he made grids.
Rows were suspects. Columns were levels of attention. Low, medium, high. Internal mention. Internal focus. Full analytical scrutiny. External exposure. Media saturation.
Then he marked outcomes.
Not deaths. That was too blunt. He marked _events_. Sudden shifts. Abrupt redirections. Leads that collapsed without explanation. Anonymous tips that arrived exactly when a line of inquiry sharpened.
The page filled faster than he expected.
He paused, pen hovering, and checked the door. No one watching. Good. Or at least nobody obvious.
Matsuda had always been bad at statistics. Numbers blurred for him, lost their authority the longer he stared. But shapes made sense. Patterns. Negative space.
And the shape forming on the page was wrong.
There were gaps where there shouldn't be gaps. High-attention suspects with no immediate consequences. Low-attention ones that triggered sudden cascades.
Except when he overlaid another variable.
_Who knew._
Not in the abstract. Not "the Task Force." Specific people. L. The Chief. Aizawa. Himself.
Matsuda circled his own name reluctantly.
Every time he had been fully briefed on a suspect before the rest of the team, the outcome deviated. Not predictably. Not repeatably. But noticeably. The system behaved… noisier.
He hated that word. It sounded like an excuse.
He flipped to a fresh page and tried again, this time pretending he didn't exist.
The chart stabilized.
Less distortion. Fewer outliers. Still strange, but within the range of coincidence. The kind of coincidence adults accepted because rejecting it required effort.
His stomach tightened.
"So I'm the glitch," he muttered.
That was when the printer across the room activated.
Matsuda jumped. Paper slid out. One page. Then another. Then stopped.
No one else reacted. Phones stayed down. L didn't look up.
Matsuda walked over slowly and picked up the sheets.
They were summaries. Case digests. The kind he was no longer supposed to receive.
He scanned the header.
_Suspect: Unconfirmed. Internal Reference Only._
The timestamp was current.
His pulse spiked.
This wasn't protocol. Someone had either made a mistake or deliberately routed the file to him. Both options were bad.
He glanced at L.
L was watching him.
Not directly. L never did anything directly. But the angle of his gaze shifted just enough that Matsuda felt seen in a way that made his skin crawl.
Permission or bait. Matsuda couldn't tell which.
He took the papers back to his desk.
This time, he didn't add them to the chart.
He waited.
Ten minutes passed.
Nothing happened.
He exhaled, almost laughed, then stopped himself. The relief felt premature, like celebrating during the opening credits.
Then his phone vibrated.
A notification from a secure channel. One he technically still had access to.
_Incident logged. Case divergence noted._
No death. No name. Just divergence.
Matsuda looked back at the printout.
The suspect hadn't been named. Not to him. Not yet.
He realized, with a cold clarity that settled behind his eyes, that the system might not care _who_ he was observing.
Only that he was.
He folded the papers and slid them into his bag.
Across the room, Aizawa was watching him now too. Suspicious. Guarded. The look you give a faulty sensor you're not ready to disconnect.
L spoke without turning. "Matsuda."
"Yes?"
"From now on, do not attempt independent analysis."
Matsuda nodded. "Understood."
L continued, "If you notice anything unusual, report it immediately."
Matsuda hesitated for a fraction of a second. Long enough to matter.
"Yes," he said.
The meeting ended soon after. Procedures updated. Controls tightened. The system congratulated itself quietly.
That night, at home, Matsuda added one final note to his chart.
A simple arrow looping back on itself.
_Observation → Adjustment → Reinforced anomaly._
He stared at it until the ink blurred.
Because the scariest part wasn't that Kira might be watching them.
It was the growing suspicion that _investigation itself_ was becoming an active participant.
And participants could be influenced.
The Task Force adjusted.
That was the problem.
They didn't announce it. No formal vote. No memo. Adjustments seeped in the way habits do. Quietly. With justification. With everyone convinced they were still acting freely.
Matsuda watched it happen from the outside.
His access was narrower now. Sanitized. He received outcomes, not processes. Conclusions, not steps. The official rationale was efficiency. Fewer leaks. Cleaner chains of custody. The unspoken rationale sat heavier.
Contain the variable.
Which meant contain him.
He stopped charting on paper. Too obvious. Too vulnerable. Instead, he memorized shapes. Where conversations slowed. Where decisions stalled. Where L hesitated a fraction longer than usual.
L never hesitated without reason.
A new suspect entered the system. Not publicly. Not even internally, not in the usual sense. No file header. No working name. Just a cluster of anomalies that didn't cohere unless you assumed intent.
Matsuda wasn't briefed.
So nothing should have happened.
Instead, the system grew erratic.
Minor things at first. A delayed warrant approval. Conflicting surveillance logs. Two analysts producing mutually exclusive summaries from the same dataset.
No deaths.
That detail mattered.
Matsuda noticed his own behavior changing. He avoided thinking about the suspect directly. When his mind drifted there, he redirected it aggressively. Counted steps. Listed baseball stats. Anything to reduce internal focus.
It felt ridiculous.
It also seemed to work.
Then L called him in.
Not to the conference room. To the observation room. The one with the one-way glass that everyone pretended not to understand.
"Matsuda," L said, crouched as always. "Sit."
Matsuda sat.
On the other side of the glass, analysts worked through data streams. None of them knew Matsuda was there. That was intentional.
"Tell me," L said, "what you are thinking about right now."
Matsuda blinked. "You want me to—"
"Answer precisely," L said.
Matsuda swallowed. "I'm trying not to think about the current suspect."
L nodded. "Why?"
"Because when I do," Matsuda said carefully, "things change."
L tilted his head. "Define change."
"Distortion," Matsuda said. "Noise. Outcomes that don't align with prior probabilities."
"So you believe your cognition affects the investigation."
"I believe _attention_ does," Matsuda replied. "Mine just seems… louder."
L was silent for a long moment.
Then he said, "We ran a test."
Matsuda's stomach dropped. "What kind of test?"
"We restricted all explicit analysis of the suspect," L said. "No names. No identifiers. No sustained focus. We monitored system stability."
"And?"
"Stability improved," L said.
Matsuda closed his eyes briefly.
"Then," L continued, "we introduced a single observer. One analyst. Fully briefed. Isolated."
Matsuda opened his eyes. "What happened to them?"
"Nothing," L said. "At first."
Matsuda waited.
"After six hours," L said, "a different suspect suffered a fatal heart attack."
Matsuda felt cold spread through his chest. "But that person wasn't—"
"Connected?" L finished. "Correct."
"So the system compensated," Matsuda said. "It redirected."
"Possibly," L replied. "Or you are pattern-matching."
Matsuda laughed once, sharply. "You don't believe that."
"I don't believe in coincidence at scale," L said. "But I also don't believe in supernatural explanations without exhaustion of alternatives."
"So what's the alternative?" Matsuda asked.
L's eyes met his. "That Kira has constructed a system that weaponizes investigation itself."
The words settled like dust.
"Then what do we do?" Matsuda asked.
"We proceed," L said. "Carefully."
"How?"
"By observing the observers," L replied.
Matsuda understood then. Fully. With a clarity that made him dizzy.
They weren't hunting Kira anymore.
They were studying feedback.
From that day on, Matsuda was doubted officially. His reports flagged as speculative. His warnings footnoted. His access restricted just enough to keep him functional, just enough to keep him relevant.
And just enough to keep him inside the loop.
The anomalies decreased.
So did the deaths.
No one called it a success.
Late one night, alone at his desk, Matsuda noticed something small. A trivial update in a closed case file. A timestamp adjustment. A correction that didn't need to exist.
He stared at it for a long time.
Then he looked away.
Nothing happened.
He didn't know whether that meant the system had stabilized.
Or whether it had learned.