Chapter 81: The Optimal Line

Park's analysis arrived at three PM Wednesday, delivered in the form of a printed document because Park trusted paper in a way she didn't trust screens—paper couldn't crash, couldn't be remotely accessed, couldn't display updated data that contradicted the data she'd already committed to presenting. She laid the document on the kitchen table with the care of a surgeon setting instruments, and the team gathered around it the way soldiers gathered around a map.

The document was titled: *Operational Parameter Optimization: Repair Rate vs. Detection Risk*.

"The detection threshold appears to be between 0.78 and 0.82 percent per session," Park said. "Below 0.78, the Arbiter shows no measurable response. Above 0.82, the probability of a pulse event approaches certainty. Between those values—" She tapped the graph on page two. A band of uncertainty, shaded gray, the mathematical territory where prediction broke down and every session was a coin flip. "Between those values, we're gambling."

"So we stay below 0.78," Ren said.

"If we stay below 0.78, the crossover point is approximately nineteen weeks. That's within our operational window—barely. But nineteen weeks of daily sessions means nineteen weeks of fragment integration, which means—" Park glanced at Mira.

"—significant composite growth," Mira finished. She sat at the end of the table, the bandage over her right eye a white flag of the damage done yesterday. Her left eye tracked the document with the concentrated precision of a person compensating for reduced perception. "At 0.75 percent repair rate per session, Leo integrates roughly one hundred and seventy fragments per night. Over nineteen weeks—one hundred and thirty-three sessions—that's approximately twenty-two thousand six hundred fragments. Total integration would reach approximately thirty thousand." She paused. "There are only ten thousand to integrate. He'll be done in approximately fifteen sessions at the current pace."

The room absorbed this. Leo registered the math the composite had already processed: 10,000 minus 7,620 equaled 2,380 fragments remaining. At 170 per session, that was 14 sessions. Two weeks. The integration would be complete long before the crossover point arrived. The channel would be fully open. The repair rate would hit its maximum.

The question was whether the maximum repair rate—achievable only after full integration—exceeded the detection threshold.

"Park," Leo said. "After full integration. What's the projected maximum repair rate?"

Park flipped to page four. A curve that started at the current 0.75-0.78 range and climbed steeply once integration passed 9,000 fragments. "Full integration opens the channel completely. No bottleneck. The feedback loop operates at theoretical maximum. The projected repair rate at full integration is—" She pointed to the number at the curve's peak. "—approximately 2.4 percent per session."

The number sat on the table.

"That's three times the detection threshold," Morrison said.

"Yes. After full integration, there's no way to repair the seal without the Arbiter knowing. The energy signature at 2.4 percent would be impossible to mask. We'd be repairing in broad daylight." Park set down the document. "Which means we have two phases. Phase one: the remaining fourteen sessions to complete integration. These sessions need to stay below the detection threshold—which they can, because the repair rate at this integration level is still below 0.78. Phase two: post-integration, maximum repair rate, sustained pulse events from the Arbiter."

"Phase two is the race," Leo said.

"Phase two is the sprint. At 2.4 percent repair, even with regular pulse events, we exceed the degradation rate. The crossover happens. But the margin depends on the Arbiter's pulse frequency and magnitude. If the Arbiter can pulse daily at Monday's magnitude, the net repair rate drops to approximately 0.6 percent—positive, but barely. If the Arbiter pulses weekly, net repair is 1.8 percent. That's meaningful progress." Park flipped to the final page. "And there's a third possibility. At full integration, Leo's channel to the seal is fully open. He's not just pushing repair energy through a partially open pipe. He's connected to the seal's entire architecture. The repair energy doesn't just flow to the east district juncture—it distributes across all seven juncture points simultaneously. The Arbiter can pulse at one point, but it can't pulse at all seven. For the first time, we'd be repairing faster than the Arbiter can target."

Kai had been listening from the doorway. He'd appeared at some point during Park's presentation—Leo hadn't noticed when, which meant the composite hadn't flagged his arrival, which meant the autonomous loop responsible for tracking household occupants had either missed the boy or had classified his arrival as a non-relevant event. Both possibilities concerned Leo for different reasons.

"The two-week gap," Kai said. Everyone turned. The boy stood in the doorframe with his laptop under one arm and the expression of a teenager who'd heard adults making plans and had identified the flaw. "Phase one takes two weeks. During those two weeks, the Arbiter might pulse anyway—not because of our sessions, but on its own schedule. Monday's first pulse happened during the day, unrelated to any session. If the Arbiter is escalating independently, we could lose another eight percent of juncture integrity before we even reach full integration."

"That's accounted for in the model," Park said.

"Is it? You modeled for one more pulse event at Monday's magnitude during the two-week window. But Monday was the first pulse we ever detected. We don't know if there were prior pulses before we had sensors in place. We don't know the historical frequency. We're modeling future behavior from two days of observation."

Park's jaw tightened. The specific tension of a scientist being challenged on her methodology by a thirteen-year-old who was right. "The model has wide uncertainty bands for exactly that reason."

"Wide enough? What if the Arbiter pulses three times in two weeks? What if it discovered it can punch and now it likes punching?" Kai walked to the table. Set his laptop down. Opened it to a graph that he'd clearly been building since last night—the same curves as Park's document but with a broader range of Arbiter behavior scenarios. "I ran Monte Carlo simulations. Ten thousand iterations with randomized pulse frequency, magnitude, and targeting. In sixty-three percent of scenarios, we reach full integration with the juncture intact. In thirty-one percent, we reach integration but the juncture has degraded below the point where full-channel repair can save it in time. And in six percent of scenarios—"

"The juncture fails before we finish," Leo said.

"During the two-week integration window." Kai's voice was flat. The delivery of a number he didn't want to be responsible for but couldn't avoid presenting because the number existed and ignoring it would be worse than facing it. "Six percent chance that the thing we're racing toward finishes after the race is already lost."

The kitchen was quiet. The refrigerator hummed. The grocery list—Sarah's handwriting, the domestic artifact of a house that was also a command center—fluttered in the draft from the ventilation. Outside, a car passed on the street. Fewer cars than a month ago. The neighborhood's pulse weakening.

"Six percent is an acceptable risk," Morrison said. "Military operations launch at ninety-percent confidence."

"Military operations don't determine whether reality continues to exist," Kai said.

Morrison looked at the boy. The look lasted three seconds—the evaluation of a general assessing a subordinate who'd just pushed back. What Morrison saw in Kai's face was what everyone saw when Kai stopped being a calculator and started being a person: the fear. Not of the number—six percent was abstract. Of the responsibility. Of being the person who'd built the model and knowing that the model was the best available and knowing that the best available wasn't good enough and knowing that the decision-makers would use his model because there was nothing else to use.

"It is an acceptable risk," Morrison said, softer. "Because the alternative is a hundred-percent chance of failure. The only way through is through."

---

Wednesday night's session was the last quiet one.

Leo sat in the chamber. Eleven stabilizers—Marcus still recovering, his projected return Thursday. The membrane at ninety-four percent. The channel at one seventy percent—slightly below the previous two nights, calibrated to Park's optimal line: the repair rate that maximized gain while staying below the detection threshold.

The fragments came.

*Death 5,041. Year five. A dragon. Not a drake—a true dragon, Class S, something he should never have engaged alone, something that should have killed him dozens of times but only killed him once because the dragon was old and slow and the fight lasted eleven minutes and the death, when it came, was fire. Not the drake's cave-fire, not the elemental's volcanic fire, but dragon fire—the kind that burned the air itself, that turned oxygen into fuel, that consumed everything between the flame and the horizon. He'd felt his skin peel. His muscles cook. His bones blacken. And underneath the pain, the craving, the terrible gratification of a drug hitting the bloodstream after a long drought—*

The composite processed the fragment. Filed it. Moved on.

Leo held the gap. A third of a second. The dragon fire's echo—not the data but the memory of heat. Real heat. The kind that cooked a man alive and left him grinning.

He held the shame. A quarter second this time. Shorter than yesterday because the composite was faster today—the autonomous loops had optimized their timing, closing the gap between processing cycles with the efficiency of a system that learned from its own operation.

The gap was narrowing. Not quickly. Not dramatically. But the trend was there. Each session made the composite faster, and each optimization reduced the space where Leo's raw consciousness could exist.

Tanaka's homework: widen the gap. The composite's homework: close it.

Two students with opposite assignments sharing the same body.

The session ended at one AM. One hundred and sixty-eight fragments. Integration: 7,788 out of 10,000. Repair rate: 0.76 percent. Below the detection threshold. Park monitored the juncture readings for an hour post-session.

No pulse.

Two nights without a pulse. The Arbiter quiet. The optimal line holding.

"It's working," Park said. Relief tempered by the scientist's constitutional allergy to premature conclusions. "Two sessions below threshold, two sessions without response. The data is consistent with the detection model."

"Two data points," Kai said from his corner. The boy had been allowed to observe tonight—Leo's decision, overriding Morrison's preference to keep Kai away from the chamber during sessions. Kai needed to see the process he'd helped design. Needed to understand what his feedback loop did to the man who used it. "Two data points is a line, not a pattern."

"A line is a start," Park said.

Leo walked upstairs. The house was dark. Sarah had left a plate of food in the microwave—the domestic persistence of a woman who fed people regardless of whether they remembered to eat, the small act of care that existed alongside the strategic calculations and the sensor readings and the death fragments. Leo ate standing up. Rice, chicken, vegetables. The composite analyzed the nutritional content and flagged a caloric deficit relative to the session's energy expenditure. Leo ignored the flag and ate because the food was warm and Sarah had made it and eating warm food that someone made for you was a human act, not a data point.

He sat in the backyard. The Adirondack chair. The night sky. The stars doing what stars did when no one needed them for navigation—existing, burning, dying across distances that made the seal's problems feel small and Leo's problems feel smaller and the Arbiter's imprisonment feel like a footnote in a universe that had bigger concerns.

The gap. A third of a second.

Stars. Cold air on his forearms. The distant sound of a dog barking three streets away—not the composite's analysis of sound frequency and estimated distance but the actual bark, the actual sound, a living thing making noise in the dark because living things made noise in the dark when they had something to say.

A third of a second became two-fifths. The composite reached. He held it. The bark. The cold. The star overhead that was probably dead already, its light traveling for centuries, arriving now, landing on his skin with the ancient patience of a photon that didn't care about seals or arbiters or committees.

Two-fifths of a second. Then the composite caught it and the stars became data and the bark became distance and the cold became temperature and the gap closed.

Wider than yesterday. The homework was working.

---

Thursday morning. Park's sensors flagged an anomaly at nine AM.

Not a pulse. Something different. The juncture readings at the east district showed a pattern that Park's algorithms hadn't been programmed to detect because the pattern hadn't existed before Tuesday. A low-frequency oscillation in the Arbiter's baseline sabotage activity—a rhythmic variation in the corrupted energy output, like a heartbeat that had been steady for two centuries and had just developed an irregularity.

Park called Leo at nine-fifteen. Her voice carried the particular energy of a researcher who'd found something unexpected—not alarm, but the heightened attention of a scientist encountering a novel data set.

"The Arbiter's baseline pattern has changed," she said. "Not increased. Changed. The sabotage energy is oscillating—cycling between higher and lower output at approximately forty-minute intervals. It started sometime overnight. I can't pinpoint the exact onset because the variation is subtle and my algorithms only flagged it during the morning data review."

"What does the oscillation mean?"

"I don't know. It could be a response to the accumulated repair energy—even below the detection threshold, four sessions' worth of repair work has produced structural changes at the juncture. The Arbiter might be probing those changes. Feeling the texture of the repair, the way you'd feel a patch on a wall to test whether the plaster had dried." Park paused. The analytical pause of a mind constructing a hypothesis from insufficient data. "Or it could be independent. The Arbiter's been operating for two hundred years. We've only been monitoring for weeks. This oscillation might be a normal behavior pattern we haven't observed before because we didn't have the instruments to detect it."

"Your read."

"My read is that the Arbiter is thinking. The pulse events were actions. This oscillation isn't an action—it's a process. Analytical. Evaluative. The Arbiter detected something—either our repair work or its own pulse events' effectiveness—and now it's assessing. Gathering information. Adjusting its model." Another pause. Longer. "Park's read, not Park's conclusion. I want that distinction noted."

Leo noted the distinction. The difference between what a scientist believed and what a scientist could prove—the gap between intuition and evidence that Park defended with the territorial aggression of a researcher who'd been trained to know the difference.

"Keep monitoring. If the oscillation changes—frequency, amplitude, targeting—I want to know immediately."

He hung up. Stood in the kitchen. Thursday morning light. Kai at the table with his rebuilt model, the Monte Carlo simulations running on his laptop with the quiet persistence of ten thousand imagined futures being computed simultaneously. Sarah at the stove, making eggs. David checking the security cameras—the routine morning assessment of a man whose job was ensuring that the house where a death counter lived remained secure against threats that ranged from Purifier assassins to nosy journalists.

Normal. The house performing its daily operations. The routines that existed because Sarah and David and Kai needed routines the way the seal needed repair pathways—structural elements that held the daily reality together while the cosmic reality eroded beneath it.

Leo's phone buzzed. Morrison.

"Chen has a meeting request from Helene."

"Sister Helene?"

"Direct. Not through Vardis. Not through the committee. Helene contacted Chen through a private academic channel—a journal editor they both know. She's requesting a meeting with Park. Specifically Park. She wants to discuss the energy signatures from the Project Anchor failures." Morrison's voice was controlled but there was something underneath the control—the heightened attention of a strategist recognizing a move he'd predicted. "She's looking for the active processes. The thing we told the committee was 'ongoing research.' Helene heard the euphemism and she's pulling the thread."

"Morrison predicted this."

"I predicted it would take months. It's been three days. Helene is faster than I expected." The compliment, from Morrison, carried the same weight as a threat assessment. "She's a physicist, not a politician. She doesn't care about Vardis's narrative strategy. She cares about the substrate interactions that killed four of her team's operations. She wants to know what redirected the energy. And she's going to Park because Park's analysis is the one that showed the redirection pattern."

"Do we say no?"

"We can. Chen can decline through diplomatic channels. But declining tells Helene that we're hiding something, which makes her look harder, which means she goes around us—to the committee, to her own research team, to independent scientists who'll start poking at the same questions." Morrison paused. "Or we say yes. Park meets Helene. They discuss the energy signatures. Park shares the data we've already submitted to the committee—nothing new, nothing classified. Helene gets her meeting. And we learn what Helene knows, what she suspects, and how close she is to finding the Arbiter on her own."

"Intelligence gathering."

"Mutual. Helene wants information. We want information about what information Helene wants. The meeting is an exchange where both sides pretend they're collaborating while actually assessing each other's position." Morrison's voice carried the dry amusement of a man who'd spent decades in rooms where everyone knew the game and played it anyway. "I'll prep Park. She needs to know what she can share and what she can't. The line is: substrate interactions, yes. Corrupted repair mechanisms, yes—in general terms. The entity behind the corruption—no. Not yet."

"Not yet."

"Not yet." Morrison hung up.

Leo set the phone down. Through the kitchen window, the street outside was quiet. Thursday morning in the east district. Fewer kids walking to school than last month. Fewer parents on the sidewalks. The slow evacuation continuing, one decision at a time, the neighborhood's population curve declining with the same patient inevitability as the seal's structural integrity.

But the tomato plant upstairs had eight leaves now. Kai's laptop ran ten thousand futures and sixty-three percent of them ended with the juncture intact. The gap between Leo's processing cycles was two-fifths of a second wide and growing. The optimal line held—below the detection threshold, above zero, the narrow band where progress was possible without triggering the intelligence that had been breaking its prison for longer than the nation had existed.

Two weeks of quiet sessions. Then full integration. Then the sprint.

Leo ate the eggs Sarah put in front of him. Tasted them—not the composite's nutritional analysis but the actual taste, the salt and the butter and the heat, a fraction of a second of unprocessed sensation slipping through the gap before the composite caught it and filed it and moved on.

A fraction of a second of eggs tasting like eggs.

He'd take it.